Sleep apnea detection method, device, equipment, medium and product

Abstract

The invention relates to the technical field of biomedical signal processing, and discloses a sleep apnea detection method, device and equipment, a medium and a product, and the method comprises the following steps: obtaining a target vital sign signal; performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level sub-band signal corresponding to the respiratory effort signal; constructing a feature matrix according to the respiratory effort signal and the wavelet level sub-band signal; the feature matrix comprises a plurality of dimensions of signal features; the signal features are obtained by performing feature extraction on the respiratory effort signal or performing relative feature extraction between sub-band signals on the wavelet level sub-band signal; inputting the feature matrix into a detection model to obtain the sleep apnea times of the target population; the detection model is obtained by training with vital sign signals as samples and signal features as labels. The sleep apnea event can be accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical signal processing, and in particular to a sleep apnea detection method, device, equipment, medium and product. Background Art

[0002] At present, in order to overcome the defect of polysomnography (PSG) in monitoring sleep quality and cardiac function of patients with Sleep Apnea-Hypopnea Syndrome (SAHS) at the same time, non-contact piezoelectric devices are used to collect vital sign aliasing signals. This method has the advantages of simple monitoring process and low monitoring complexity. At the same time, the comfort of the subjects is improved, which can meet the needs of long-term home use.

[0003] However, there are still some problems with non-contact monitoring solutions. On the one hand, the signals collected by the piezoelectric system include cardiac impact and respiratory information, while the energy of the cardiac impact signal is weaker than that of the respiratory signal, and its anti-interference ability is poor. Its morphology will be affected by the amplitude increase of respiratory activity. Moreover, during sleep, the respiratory frequency will change suddenly due to the influence of pathological factors of the respiratory system. On the other hand, the respiratory rhythm of different individuals is different, and the respiratory intensity also varies. If the fixed frequency filtering method is used to extract the respiratory effort signal, too high a cutoff frequency will contain too much high-frequency noise, and too low a cutoff frequency will cause signal distortion. In addition, the currently commonly used feature description method changes the signal strength due to the change in the sensor sensing position after body movement and sleeping posture changes. At this time, the comparison with the signal fragment of the previous time is not compared at the same baseline signal strength. The change in the current signal baseline strength will cause deviations in feature expression.

[0004] Therefore, there is an urgent need for a sleep apnea detection method that can overcome the above-mentioned defects and can achieve accurate detection of sleep apnea events. Summary of the invention

[0005] In view of this, the present application provides a sleep apnea detection method, device, equipment, medium and product, which can accurately detect sleep apnea events. The technical solution is as follows.

[0006] In a first aspect, the present invention provides a sleep apnea detection method, the method comprising:

[0007] Acquire target vital sign signals;

[0008] Performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal;

[0009] According to the respiratory effort signal and the wavelet hierarchical subband signal, a feature matrix is ​​constructed; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features from the respiratory effort signal or extracting relative features between subband signals from the wavelet hierarchical subband signal;

[0010] The feature matrix is ​​input into the detection model to obtain the number of sleep apnea times; the detection model is trained using vital sign signals as samples and signal features as annotations.

[0011] In an optional implementation, the acquiring of a target vital sign signal includes:

[0012] Obtain original vital signs signals of the target population;

[0013] According to a set time period, the original vital sign signal is divided into a time period set of a plurality of signal segments;

[0014] Get the amplitude standard deviation of each time period in the time period set;

[0015] Determine the invalid signal segment in the signal segment according to the magnitude relationship between the amplitude standard deviation and the amplitude threshold;

[0016] An invalid mark is added to the invalid signal segment to obtain the target vital sign signal.

[0017] In an optional implementation, the target vital sign signal is subjected to wavelet transformation to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal, including:

[0018] According to the set time period, the invalid signal is avoided, and the original vital sign signal is segmented and standardized to obtain standardized signal segments;

[0019] The standardized signal segment is decomposed and reconstructed using wavelet transform to obtain subband signals at different wavelet levels; the subband signals include approximate subband signals and detail subband signals;

[0020] The similar subband signals and detail subband signals of different wavelet decomposition levels are screened by using the Pearson correlation coefficient to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal.

[0021] In an optional implementation, constructing a feature matrix according to the respiratory effort signal and the wavelet level subband signal includes:

[0022] Performing feature extraction on the respiratory effort signal to extract variance features, absolute mean features, root mean square features, respiratory power features, and respiratory frequency features;

[0023] Comparing the approximate subband signal and the detail subband signal of the wavelet level subband signal, extracting the relative features of the Pearson coefficient, the relative features of the average Euclidean distance and the relative features of the subband variance ratio between the approximate subband signal and the detail subband signal;

[0024] The standardized signal segments are divided according to the extracted features and relative features to obtain the feature matrix.

[0025] In an optional implementation, the feature matrix is ​​input into a detection model to obtain the number of sleep apnea, including:

[0026] The feature matrix is ​​input into the trained detection model to obtain the detection results of each standardized signal segment;

[0027] The number of sleep apnea events of the target population is obtained according to the detection results of each standardized signal segment.

[0028] In an optional embodiment, the detection model is obtained by training a preset extreme gradient boosting machine learning classification model.

[0029] The sleep apnea detection method provided by the present invention has the following advantages.

[0030] The sleep apnea detection method of the present invention first obtains the target vital sign signal (piezoelectric aliasing signal) to be checked, that is, the clinical vital sign signal of the target population, and performs preprocessing to eliminate body movement and bed leaving segments and obtain a valid signal segment set for detection. Subsequently, the approximate coefficients and detail coefficients at different resolutions (i.e., wavelet transform levels) are obtained by wavelet decomposition, and the correlation of coefficients at different levels is compared in combination with Pearson correlation analysis, thereby screening the respiratory effort signal and the corresponding wavelet level subband signal. After that, the respiratory effort signal is feature extracted, and the relative feature extraction between the subband signals of the wavelet level subband signal is performed to construct a feature matrix. Finally, the feature matrix is ​​input into the detection model to obtain the number of sleep apnea of ​​the target population; the detection model is obtained by training the machine learning classification model with the vital sign signal as the sample and the signal feature as the annotation. The sleep apnea detection method of the present invention extracts the respiratory effort signal reflecting the chest and abdomen fluctuation from the piezoelectric aliasing signal, and the sleep apnea is judged based on the respiratory effort signal. The robustness of the detection signal is improved, and the complexity of feature extraction caused by signal interference is reduced. Taking advantage of the multi-resolution characteristics of wavelet transform in the time-frequency domain, the piezoelectric aliasing signal is subjected to wavelet transform, and multiple sub-band signals corresponding to the key frequency interval of breathing are reconstructed according to the approximate coefficients and detail coefficients of wavelet decomposition as candidate signals of respiratory effort. The Pearson correlation coefficients of the sub-band signals obtained at adjacent levels (reconstructed according to the approximate coefficients and detail coefficients, respectively) are calculated, and the sub-band signals corresponding to respiratory effort are selected by comparing the correlation coefficients of different sub-band signals. A fragment feature based on the comparison of multi-sub-band (resolution) signals is proposed, combined with the volatility features extracted from the respiratory effort signal, to extract features from the multi-resolution sub-band signals and fuse them, and the fused feature set is used for sleep apnea detection.

[0031] In a second aspect, the present invention provides a sleep apnea detection device, the device comprising:

[0032] An acquisition module, used for acquiring a target vital sign signal;

[0033] A reconstruction module, used for performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal;

[0034] An extraction module constructs a feature matrix according to the respiratory effort signal and the wavelet hierarchical subband signal; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features from the respiratory effort signal or extracting relative features between subband signals from the wavelet hierarchical subband signal;

[0035] The detection module is used to input the feature matrix into the detection model to obtain the number of sleep apnea; the detection model is trained using vital sign signals as samples and signal features as annotations.

[0036] In an optional implementation, the acquisition module is specifically used to:

[0037] Obtain original vital signs signals of the target population;

[0038] According to a set time period, the original vital sign signal is divided into a time period set of a plurality of signal segments;

[0039] Get the amplitude standard deviation of each time period in the time period set;

[0040] Determine the invalid signal segment in the signal segment according to the magnitude relationship between the amplitude standard deviation and the amplitude threshold;

[0041] An invalid mark is added to the invalid signal segment to obtain the target vital sign signal.

[0042] In an optional implementation, the reconstruction module is specifically used to:

[0043] According to the set time period, the invalid signal is avoided, and the original vital sign signal is segmented and standardized to obtain standardized signal segments;

[0044] The standardized signal segment is decomposed and reconstructed using wavelet transform to obtain subband signals at different wavelet levels; the subband signals include approximate subband signals and detail subband signals;

[0045] The similar subband signals and detail subband signals of different wavelet decomposition levels are screened by using the Pearson correlation coefficient to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal.

[0046] In an optional implementation, the extraction module is specifically used to:

[0047] Performing feature extraction on the respiratory effort signal to extract variance features, absolute mean features, root mean square features, respiratory power features, and respiratory frequency features;

[0048] Comparing the approximate subband signal and the detail subband signal of the wavelet level subband signal, extracting the relative features of the Pearson coefficient, the relative features of the average Euclidean distance and the relative features of the subband variance ratio between the approximate subband signal and the detail subband signal;

[0049] The standardized signal segments are divided according to the extracted features and relative features to obtain the feature matrix.

[0050] In an optional implementation, the detection module is specifically used to:

[0051] The feature matrix is ​​input into the trained detection model to obtain the detection results of each standardized signal segment;

[0052] The number of sleep apnea events of the target population is obtained according to the detection results of each standardized signal segment.

[0053] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the sleep apnea detection method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0054] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the sleep apnea detection method of the first aspect or any corresponding embodiment thereof.

[0055] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, wherein the computer instructions are used to enable a computer to execute the sleep apnea detection method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 The figure is a flowchart of a sleep apnea detection method according to an exemplary embodiment.

[0058] Figure 2 It is a specific flowchart of a sleep apnea detection method according to an exemplary embodiment.

[0059] Figure 3 is a schematic diagram of a 60-second signal of normal breathing according to an exemplary embodiment.

[0060] Figure 4 is a schematic diagram of a 60-second signal including respiratory arrest according to an exemplary embodiment.

[0061] Figure 5It is a structural schematic diagram of a sleep apnea detection device provided in an embodiment of the present application.

[0062] Figure 6 It is a structural schematic diagram of a computer device provided by an optional embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0064] It should be understood that the "indication" mentioned in the embodiments of the present application can be a direct indication, an indirect indication, or an indication of an association relationship. For example, A indicates B, which can mean that A directly indicates B, for example, B can be obtained through A; it can also mean that A indirectly indicates B, for example, A indicates C, and B can be obtained through C; it can also mean that there is an association relationship between A and B.

[0065] In the description of the embodiments of the present application, the term "corresponding" may indicate a direct or indirect correspondence between two items, or an association relationship between the two items, or a relationship between indication and being indicated, configuration and being configured, and the like.

[0066] In an embodiment of the present application, "predefinition" can be achieved by pre-saving corresponding codes, tables or other methods that can be used to indicate relevant information in a device (for example, including a terminal device and a network device). The present application does not limit its specific implementation method.

[0067] At present, the standardized program for monitoring sleep quality and cardiac function of patients with sleep apnea-hypopnea syndrome (SAHS) is polysomnography (PSG), which can obtain physiological information of the human body through bioelectricity of different parts of the human body or through different sensors, including electroencephalogram (EEG), electrocardiogram (ECG), electromyogram (EMG), electrooculogram (EOG), blood oxygen saturation (SaO2), nasal airflow temperature / pressure (Flow-T / P), etc., among which electrocardiogram is the gold standard for clinical detection and diagnosis of cardiovascular diseases. However, the polysomnography method has the disadvantages of complex operation, strong restraint, high price, high monitoring environment requirements, huge human resource consumption, and the need for subjects to stick electrodes on the body surface for a long time, which may affect sleep quality and even cause allergic reactions, which greatly limits its wide application in the monitoring, diagnosis and treatment of sleep apnea syndrome and cardiovascular diseases. In addition, this method relies on the use of professional equipment and professionals in medical institutions to analyze signals, and generally requires monitoring in medical institutions, and it is impossible to conduct long-term continuous monitoring of patients at home.

[0068] The use of non-contact piezoelectric devices to collect vital sign aliasing signals has the advantages of simple monitoring process and low monitoring complexity. At the same time, the comfort of the subjects is improved, which can meet the needs of long-term home use. For example, patent CN102641125A discloses a non-contact sleep apnea determination device, which detects sleep apnea by emitting microwave reflection waves from the chest and abdomen of the subject during sleep. However, monitoring using microwave reflection waves will be affected by environmental conditions or other electromagnetic waves. In addition, such as patent CN204318758U, a sleep breathing signal acquisition device based on an air mattress, the invention uses a piezoelectric sensing mattress to collect breathing signals for sleep apnea detection; piezoelectric-based detection technology can effectively reduce sleep interference, but the portability of the mattress still has room for improvement. For example, patents CN111035367B, a non-intrusive sleep apnea method and system, and CN116401598A, a sleep apnea detection method and device based on physiological signals, proposed a sleep apnea detection device based on a suboccipital piezoelectric sensor. The former proposed a sleep apnea detection method based on aliased signal characteristics, and the latter completed sleep apnea detection by extracting the characteristics of the complete cardiac impact signal and the low-frequency breathing baseline. Reference [1] uses piezoelectric sensing to achieve non-intrusive monitoring, and some of the features used are described by ratio between the minimum and maximum values, aiming to characterize the signal amplitude change by describing the volatility characteristics of the low-frequency component of the signal with a fixed scale, thereby completing apnea detection. At present, there are still some problems with the relevant schemes, including:

[0069] On the one hand, the signal collected by the piezoelectric system is an aliased signal, which can collect body surface vibration information during sleep, including cardiac impact and respiratory information. The correlation between cardiac impact and respiratory events is mainly reflected in heart rate variability, while the energy of cardiac impact signal is weaker than that of respiratory signal, and its anti-interference ability is poor. Its morphology will be affected by the amplitude increase of respiratory activity, so it is challenging to locate the cardiac impact waveform and estimate the heart rate. The respiratory effort signal can directly reflect the respiratory activity of the chest and abdomen. In the clinical detection standard of apnea, it can be used for home monitoring of adults. Moreover, the respiratory rate in clinical research will change with the physiological state of the subject, especially in patients, it is not a steady fluctuation. During sleep, the respiratory rate will mutate due to the influence of respiratory pathological factors. Therefore, the respiratory effort extraction of clinical data requires the analysis of the time-frequency domain changes of the signal, and the respiratory extraction is performed through the signal processing method that takes into account the time-frequency domain changes. On the other hand, in practical applications, different individuals have different physiological factors, their respiratory rhythms are different, and there are also differences in respiratory intensity. At present, the fixed frequency filtering method extracts the respiratory effort signal. If the cut-off frequency is too high, it will contain too much high-frequency noise, and if the cut-off frequency is too low, it will cause signal distortion. Therefore, in clinical applications, it is necessary to select a suitable frequency interval for individual respiration reconstruction to better obtain a smooth and accurate respiratory signal. In addition, the currently commonly used feature description method is to perform absolute calculations on the signal fragments to describe the signal morphology, or to compare and calculate with the preceding signal to obtain the fluctuation change characteristics (CN117017233B). However, after the occurrence of body movement and sleeping posture changes, the signal strength also changes due to the change in sensor sensing position. At this time, it is compared with the signal fragment of the preceding time, not at the same reference signal strength. The change in the current signal reference strength will cause a deviation in feature expression.

[0070] Therefore, in order to solve the above problems, an embodiment of the present invention provides a sleep apnea detection method, which can achieve accurate detection of sleep apnea events.

[0071] The method flow of the sleep apnea detection method of the present invention is as follows Figure 1 As shown, the following steps are included.

[0072] S101. Acquire target vital sign signals.

[0073] Specifically, the vital sign signals of clinical subjects are obtained in a non-intrusive manner, that is, the clinical vital sign signals of sleep apnea patients and healthy people are obtained through non-intrusive sensor acquisition equipment, and the signals can be collected for a whole night for analysis.

[0074] Optionally, the collected clinical vital signs signals may include signals of body movement status and bed leaving status, so the original vital signs signals need to be preprocessed, the collected clinical vital signs signals are divided into several signal segments according to set time periods, and the signal segments containing body movement status and bed leaving status are eliminated, so as to obtain a set of valid signal segments that can be used for detection.

[0075] S102 , performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal.

[0076] Specifically, before performing the wavelet transformation, the target vital sign signal needs to be standardized to remove the noise. Subsequently, the standardized signal segment is decomposed using the wavelet transform. Through multi-level wavelet decomposition, approximate coefficients and detail coefficients of different resolutions can be obtained, and subband signals of different frequency domain ranges can be reconstructed. Combined with the conventional respiratory frequency (within 0.7Hz) and the wavelet decomposition frequency range calculation rules, the number of wavelet decomposition layers and the number of wavelet reconstruction layers of focus are set to obtain the candidate subband signal of respiratory effort. The subband signal reconstructed by the approximate coefficient is called the approximate subband signal, and the subband signal reconstructed by the detail coefficient is called the detail subband signal. The Pearson correlation coefficients pa_j and pd_j of the approximate subband signal A_j and the detail subband signal D_j of the current wavelet decomposition level (level j) and the approximate subband signal A_j-1 of the previous level are calculated respectively. By comparing the Pearson correlation coefficient, the detail subband signal D_j of the current level with a higher correlation with the approximate subband coefficient of the previous level (i.e., satisfying pd_j>pa_j) is selected as the breathing effort signal. Because in the wavelet decomposition of the lower level, the frequency boundary between the high frequency and low frequency division is higher, breathing and part of the noise will be divided into low-frequency components. When the decomposition is sufficient, the breathing will be divided into the high-frequency components of the corresponding level. Therefore, the correlation of the high and low frequency subband signals (corresponding to the detail coefficient reconstruction signal and the approximate coefficient reconstruction signal, respectively) is compared. The detail coefficient reconstruction signal containing breathing at the level with sufficient decomposition has a higher correlation with the approximate coefficient reconstruction signal of the previous level. Thus, the breathing effort signal and the wavelet level subband signal corresponding to the breathing effort signal are obtained.

[0077] S103, constructing a feature matrix according to the respiratory effort signal and the wavelet level subband signal; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features from the respiratory effort signal or extracting relative features between subband signals from the wavelet level subband signal.

[0078] Specifically, the variance, absolute mean, root mean square, respiratory power, and respiratory frequency features are extracted from the respiratory effort signal, and the two sub-band signals at the same level are compared to extract the Pearson coefficient, average Euclidean distance, and sub-band variance ratio to sub-band relative features between the two. According to the extracted features and relative features, the above-mentioned target vital sign signal (signal for a whole night) is processed and divided into M segments, that is, each subject (signal for a whole night) corresponds to an 8×M segment feature matrix.

[0079] S104, inputting the feature matrix into a detection model to obtain the number of sleep apnea times; the detection model is trained using vital sign signals as samples and signal features as annotations.

[0080] Specifically, the detection model is trained based on a machine learning classification model. The feature matrix is ​​classified through the machine learning classification model to obtain the detection results of each standardized signal segment. Based on the detection results of each standardized signal segment, the number of sleep apnea episodes of the target population throughout the night can be obtained.

[0081] In order to better illustrate the sleep apnea detection method provided in the above embodiment, a specific example will be used for detailed description below. Figure 2 As shown, the following steps are included.

[0082] First, the piezoelectric aliasing signal is collected. In this example, a non-disturbance sensor signal acquisition device is used to collect the piezoelectric aliasing signal of clinical patients in real time. The non-disturbance sensor signal acquisition device mainly consists of two parts: a signal acquisition module and a data storage module. The signal acquisition module is placed under the pillow, and the clinical patient lies on his back on the pillow. The patient's body generates micro-vibrations due to the heart pumping blood to form a force signal. The piezoelectric sensor with a sampling rate of 1000hz in the signal acquisition module is placed under the pillow to collect the vital signs signals (i.e., piezoelectric aliasing signals) of clinical subjects (including apnea patients and healthy people) for about 7 hours throughout the night. The signal acquisition module can convert the force signal into an analog electrical signal, and then filter, amplify, and A / D convert the analog electrical signal into a digital signal with a sampling rate of 1000Hz through the built-in filtering circuit, amplifying circuit, and A / D conversion circuit to obtain a piezoelectric aliasing signal, which is the vital signs signal used in the above embodiment.

[0083] After the piezoelectric aliasing signal is collected, the signal needs to be preprocessed. Signal preprocessing mainly includes the specific location marking of invalid signals (body movement and leaving the bed), DC and power frequency removal, etc. The body movement judgment method based on signal energy can be used to eliminate invalid (body movement, leaving the bed) segments.

[0084] Specifically, the collected piezoelectric aliasing signal is divided into 60s signal segments, and then 2s is set as a time window, and the 60s signal segment is further divided into 30 time period sets S(n) with 2s as the time window. The amplitude standard deviation of each time period in the time period set S(n) is calculated. The standard deviation set SD(n) = [SD0, SD1,,…,SD N-1 ] Where n = 0, 1, ..., N-1, calculate the median of the standard deviation set SD(n) to get Median_SD. Then, set the upper and lower thresholds threshold1 and threshold2 of the 60s signal segment. If the energy of the 2s segment is greater than the upper threshold, it is identified as a body motion state. That is, if SD i If the value of is greater than (Median_SD×threshold1), it is judged as a body movement state; if the energy of the 2s segment is less than the lower threshold, it is identified as a bed leaving state. i If the value is less than (Median_SD×threshold2), it is determined to be in the bed leaving state. To ensure the reliability of subsequent signal analysis and feature calculation, the specific location and duration of the body movement (from the start to the end of a body movement) will be marked with an accuracy of 2s, and the location of bed leaving will be marked similarly. The 2s signal is marked as body movement state or bed leaving state, and the body movement and bed leaving events obtained with the start and end time are uniformly classified as invalid signals, called invalid sequences. Figure 3 and Figure 4 , Figure 3 This is a schematic diagram of a 60-second signal of normal breathing. Figure 4 This is a schematic diagram of a 60-second signal that includes a respiratory pause (during the red box).

[0085] After the preprocessing is completed and the invalid sequence is obtained, the multi-resolution signal based on wavelet transform and the Pearson correlation coefficient are used to screen out the respiratory effort signal and the corresponding wavelet level. The goal of this step is to select the respiratory effort signal and the corresponding wavelet level subband signal from the reconstructed signals of different resolutions. It includes the following process.

[0086] According to the position of the invalid sequence mentioned above, the invalid signal marked above is avoided and the 60s segmentation is performed again to obtain clean 60s signal segments. Specifically, the piezoelectric aliasing signal of the whole night is segmented into 60s segments, and adjacent segments do not overlap. When the second half of the segment (the last 30s) contains body movement, it moves forward and retreats to the end of the segment to perform segmentation before the start of the invalid sequence, and the next segment starts to be segmented after the end of the invalid sequence to ensure that the invalid sequence will not be divided into the signal segment. Thus, the whole night signal is segmented to obtain N 60s piezoelectric aliasing signal segments, none of which contain body movement and bed leaving status.

[0087] Since there may be significant amplitude differences in the physiological signals collected from different individuals, it is necessary to perform Z-score normalization on the clean 60s signal segments obtained by re-segmentation to reduce errors. The expression for Z-score normalization is:

[0088]

[0089] In the formula, mean(X(n)) represents the mean value of the signal, and std(X(n)) represents the standard deviation of the signal.

[0090] The normalized signal segments are passed through a fourth-order Butterworth low-pass filter with a cut-off frequency of 10Hz to remove noise and obtain the filtered signal.

[0091] Subsequently, wavelet transform is used to decompose the normalized signal segments. Through multi-level wavelet decomposition, approximate coefficients and detail coefficients with different resolutions can be obtained, and sub-band signals in different frequency ranges can be reconstructed. Specifically, for the 60s normalized segment obtained after 10Hz low-pass filtering, the wavelet basis of "db4" is used as a template, and the signal segment is wavelet decomposed to 11 layers. Considering the range of daily breathing frequencies and the signal sampling rate (1000Hz), sub-band signals of the 9th, 10th, and 11th layers are reconstructed according to the detail coefficients and approximate coefficients respectively. Specifically, the sub-band signals reconstructed from the approximate coefficients and detail coefficients of the 9th layer are denoted as A_9 and D_9 respectively; similarly, the 10th and 11th layers are denoted as A_10, D_10 and A_11, D_11 respectively. The sub-band signals obtained by wavelet decomposition and reconstruction have the same length of 60s as the signal segment before processing.

[0092] Next, appropriate sub-band signals are selected as respiratory effort signals through the Pearson correlation coefficients between sub-bands. Calculate the correlation coefficient pa_10 between A_9 and A_10, and the correlation coefficient pd_10 between A_9 and D_10; similarly, calculate the correlation coefficients between A_10 and A_11 and D_11 as pa_11 and pd_11 respectively. By comparison, select the detail coefficient reconstructed signal of this level with a higher correlation coefficient with the approximate coefficient reconstructed signal of the previous level as the respiratory effort signal (for example, if pa_11 < pd11, then select D_11 as the respiratory effort). If there is no correlation coefficient that meets the conditions, D_9 is default selected as the respiratory effort signal according to experience. At the same time, the level where the respiratory effort signal is located is selected as the wavelet level corresponding to respiration, and the two sub-band signals will be used in the subsequent feature extraction steps.

[0093] After obtaining the respiratory effort signal and the corresponding wavelet level subband signal, the features of the respiratory effort signal will be extracted, and the relative features between the two subband signals of the wavelet level corresponding to the respiratory effort signal will be extracted. Four features, namely variance, absolute mean, root mean square, respiratory power, and respiratory frequency, are extracted from the selected respiratory effort signal, and the two subband signals of the same level are compared to extract three more subband relative features, namely the Pearson coefficient, average Euclidean distance, and subband variance ratio, for a total of 8 features.

[0094] The feature extraction of the respiratory effort signal specifically includes the following processes.

[0095] The variance calculation formula is used to quantify the fluctuation characteristics of the signal segment and extract the variance of the respiratory effort signal. The formula is as follows:

[0096]

[0097] Where μ is the mean value of the 10s signal segment, x i (i=1,2…,N) is a 10 s respiratory effort signal segment with a sampling rate of 1000 Hz.

[0098] The IQR (Interquartile Range) value of the signal amplitude sequence X(n) is calculated using the interquartile range formula to extract the absolute mean of the respiratory effort signal. The formula is as follows:

[0099] IQR = Q3[X(n)]-Q1[X(n)];

[0100] In the formula, Q3 is the third quartile and Q1 is the first quartile.

[0101] The root mean square value of the signal X(n) is calculated by the root mean square calculation formula, and the root mean square of the breathing effort signal is extracted. The formula is as follows:

[0102]

[0103] The 10-second respiratory effort signal is subjected to discrete Fourier transform to obtain a power spectrum. The spectrum components from 0 to 0.7 Hz are summed to extract the respiratory power of the respiratory effort signal. The formula is as follows:

[0104]

[0105] In the formula, f i is the frequency, and according to the frequency resolution, 0-0.7 Hz is divided into K-1 intervals, and the i-th frequency interval corresponds to the power spectrum P(f i ). Sum to obtain the breathing power P of the signal resp .

[0106] The 10-second respiratory effort signal is subjected to discrete Fourier transform to obtain a spectrum, and the frequency corresponding to the maximum value of the power spectrum is selected as the respiratory frequency of the signal segment to extract the respiratory frequency of the respiratory effort signal.

[0107] In the selected breathing effort level, the detail coefficient reconstructed subband signal and the approximate coefficient reconstructed signal are compared to extract relative features of the two, which specifically includes the following process.

[0108] Calculate the Pearson correlation coefficient between two 10s sub-band signals. The formula is as follows:

[0109]

[0110] Where A(n) (n=1, 2, 3, ..., N) is the subband signal reconstructed by the approximate coefficients of the 10s segment of the selected level, and D(n) (n=1, 2, 3, ..., N) is the subband signal reconstructed by the corresponding detail coefficients. and are the signal means of the above two sub-band signals respectively.

[0111] Since there is a difference in signal energy loss after wavelet decomposition, the two sub-band signals are first normalized separately, and the expression is:

[0112]

[0113] Wherein, min(A(n)) refers to the minimum value of signal A(n), and max(A(n)) refers to the maximum value of the signal. Similarly, the reconstructed signal D(n) of the level detail coefficient is also normalized by the normalization formula.

[0114] The Euclidean distance of the normalized signal is calculated as follows:

[0115]

[0116] Calculate the variance ratio of the two sub-band signals, the expression is:

[0117]

[0118] Where A(n) (n=1, 2, 3, ..., N) is the subband signal reconstructed by the approximate coefficients of the 10s segment of the selected level, and D(n) (n=1, 2, 3, ..., N) is the subband signal reconstructed by the corresponding detail coefficients. and are the signal means of the above two sub-band signals respectively.

[0119] After extracting features from the respiratory effort signal and extracting relative features between the two sub-band signals at the wavelet level corresponding to the breathing, sleep apnea can be detected, which specifically includes the following process.

[0120] The whole night signal is processed and finally divided into M 10s segments, that is, each subject (a whole night signal) corresponds to an 8×M segment feature matrix. When the current segment contains a respiratory event, the segment is marked as a positive sample "1", and if it does not contain a respiratory event, it is marked as a negative sample "0". Each set of features (i.e., the following formula F(i), i=1,2,…,8) of the 8×M feature matrix of each subject is standardized.

[0121]

[0122] Select XGboost as the classifier for apnea detection, and set the initialization parameters of the classifier. Divide the subjects into two groups: training group and test group, with a ratio of 8:2. Optimize the classifier parameters through five-fold cross validation and grid search, and determine the training parameters. Train the XGboost classifier with the data of the training group.

[0123] After the classifier training is completed, the data of the test group is obtained and imported into the trained XGboost classifier. Based on a large amount of data experience, the prediction results of all 10s segments of the test subjects are counted. The segment prediction decision is "1", which means that it is predicted to be a signal segment where an apnea event occurs; the prediction decision is "0", which means a normal breathing signal. In this way, the prediction result of each segment is obtained, that is, the signal can be judged with an accuracy of 10s whether there is a respiratory event, and the result is output. All 10s segments of each subject are restored and sorted in the order before segmentation, that is, each subject corresponds to a column of prediction values ​​composed of "0" and "1" with a front-to-back order. Two or more consecutive positive samples "1" are retained as one "1", and independent "1" remains unchanged, that is, the respiratory events that were separated due to segmentation are restored and fused.

[0124] Finally, the number of "1"s for each subject is counted, that is, the predicted number of apnea events, that is, the start of apnea to the recovery of breathing is counted as one time. According to the number of Events obtained and the total time TST (in hours) of the original piezoelectric aliasing signal of the subject, the AHI (Apnea-Hypopnea Index) is calculated, and the expression is:

[0125]

[0126] The subjects' sleep apnea severity was divided into the following four categories based on the calculated AHI:

[0127] Healthy (normal): AHI < 5. Within this range, it means that the subject has very few apneas and hypopneas, which is generally considered a normal sleep pattern.

[0128] Mild patients: 5≤AHI<15. Within this range, the subjects may have some sleep apnea or hypopnea, but these phenomena are not enough to seriously affect the sleep quality and may have mild symptoms.

[0129] Moderate patients: 15≤AHI<30. Within this range, the subjects have more frequent sleep apnea or hypopnea, which may affect the quality of sleep, usually accompanied by a certain degree of daytime sleepiness or fatigue.

[0130] Severe patients: AHI ≥ 30. Subjects within this range experience more severe sleep apnea or hypopnea, which usually has a significant impact on health and may experience severe drowsiness, increased cardiovascular risk and other problems.

[0131] In summary, the sleep apnea detection method provided by the embodiment of the present invention first obtains the target vital sign signal (piezoelectric aliasing signal) to be checked, that is, the clinical vital sign signal of the target population, and pre-processes to eliminate body movement and bed leaving segments and obtains a valid signal segment set for detection. Subsequently, the approximate coefficients and detail coefficients at different resolutions (i.e., wavelet transform levels) are obtained by wavelet decomposition, and the correlation of coefficients at different levels is compared in combination with Pearson correlation analysis, thereby screening the respiratory effort signal and the corresponding wavelet level subband signal. After that, the respiratory effort signal is feature extracted, and the relative feature extraction between the subband signals of the wavelet level subband signal is performed to construct a feature matrix. Finally, the feature matrix is ​​input into the detection model to obtain the number of sleep apnea of ​​the target population; the detection model is obtained by training the machine learning classification model with the vital sign signal as the sample and the signal feature as the annotation. The sleep apnea detection method of the present invention extracts the respiratory effort signal reflecting the chest and abdominal fluctuations from the piezoelectric aliasing signal, and makes a judgment on sleep apnea based on the respiratory effort signal. The robustness of the detection signal is improved, and the complexity of feature extraction caused by signal interference is reduced. With the multi-resolution characteristics of wavelet transform in the time-frequency domain, the piezoelectric aliasing signal is subjected to wavelet transform, and multiple sub-band signals corresponding to the key frequency interval of breathing are reconstructed according to the approximate coefficients and detail coefficients of wavelet decomposition as candidate signals of respiratory effort. The Pearson correlation coefficients of the sub-band signals obtained at adjacent levels (reconstructed according to the approximate coefficients and detail coefficients, respectively) are calculated, and the sub-band signals corresponding to respiratory effort are selected by comparing the correlation coefficients of different sub-band signals. A fragment feature based on the comparison of multi-sub-band (resolution) signals is proposed, combined with the volatility features extracted from the respiratory effort signal, to extract features from multi-resolution sub-band signals and perform fusion, and the fused feature set is used for sleep apnea detection.

[0132] In the embodiments of the present application, a sleep apnea detection device is also provided, which is used to implement the above embodiments and preferred implementations, and will not be repeated here. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0133] The present application provides a sleep apnea detection device. Figure 5 : is a schematic diagram of a sleep apnea detection device provided in an embodiment of the present application, the device comprising:

[0134] An acquisition module 501 is used to acquire a target vital sign signal;

[0135] A reconstruction module 502 is used to perform wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal;

[0136] The extraction module 503 constructs a feature matrix according to the respiratory effort signal and the wavelet hierarchical subband signal; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features from the respiratory effort signal or extracting relative features between subband signals from the wavelet hierarchical subband signal;

[0137] The detection module 504 is used to input the feature matrix into the detection model to obtain the number of sleep apnea; the detection model is trained using vital sign signals as samples and signal features as annotations.

[0138] In an optional implementation, the acquisition module 501 is specifically configured to:

[0139] Obtain original vital signs signals of the target population;

[0140] According to a set time period, the original vital sign signal is divided into a time period set of a plurality of signal segments;

[0141] Get the amplitude standard deviation of each time period in the time period set;

[0142] Determine the invalid signal segment in the signal segment according to the magnitude relationship between the amplitude standard deviation and the amplitude threshold;

[0143] An invalid mark is added to the invalid signal segment to obtain the target vital sign signal.

[0144] In an optional implementation, the reconstruction module 502 is specifically configured to:

[0145] According to the set time period, the invalid signal is avoided, and the original vital sign signal is segmented and standardized to obtain standardized signal segments;

[0146] The standardized signal segment is decomposed and reconstructed using wavelet transform to obtain subband signals at different wavelet levels; the subband signals include approximate subband signals and detail subband signals;

[0147] The similar subband signals and detail subband signals of different wavelet decomposition levels are screened by using the Pearson correlation coefficient to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal.

[0148] In an optional implementation, the extraction module 503 is specifically configured to:

[0149] Performing feature extraction on the respiratory effort signal to extract variance features, absolute mean features, root mean square features, respiratory power features, and respiratory frequency features;

[0150] Comparing the approximate subband signal and the detail subband signal of the wavelet level subband signal, extracting the relative features of the Pearson coefficient, the relative features of the average Euclidean distance and the relative features of the subband variance ratio between the approximate subband signal and the detail subband signal;

[0151] The standardized signal segments are divided according to the extracted features and relative features to obtain the feature matrix.

[0152] In an optional implementation, the detection module 504 is specifically configured to:

[0153] The feature matrix is ​​input into the trained detection model to obtain the detection results of each standardized signal segment;

[0154] The number of sleep apnea events of the target population is obtained according to the detection results of each standardized signal segment.

[0155] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0156] The sleep apnea detection device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0157] The embodiment of the present invention also provides a computer device having the above Figure 5 The sleep apnea detection device shown.

[0158] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphic information in a graphical user interface on an external input / output device (such as a display device coupled to an interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0159] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0160] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0161] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0162] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0163] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0164] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0165] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0166] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A sleep apnea detection method, characterized in that: The method comprises: Acquire target vital sign signals; Performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal; According to the respiratory effort signal and the wavelet level sub-band signal, a feature matrix is ​​constructed; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features of the respiratory effort signal or extracting relative features between sub-band signals of the wavelet level sub-band signal; The feature matrix is ​​input into a detection model to obtain the number of sleep apnea times; the detection model is trained using vital sign signals as samples and signal features as annotations.

2. The method according to claim 1, characterized in that The step of acquiring a target vital sign signal comprises: Obtain original vital signs signals of the target population; According to a set time period, the original vital sign signal is divided into a time period set of a plurality of signal segments; Obtaining the amplitude standard deviation of each time period in the time period set; Determining invalid signal segments in the signal segments according to a magnitude relationship between the amplitude standard deviation and the amplitude threshold; An invalid mark is added to the invalid signal segment to obtain the target vital sign signal.

3. The method according to claim 2, characterized in that The step of performing a wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal includes: According to the set time period, avoiding the invalid signal, segmenting and standardizing the original vital sign signal to obtain standardized signal segments; Decomposing and reconstructing the standardized signal segment by using wavelet transform to obtain subband signals of different wavelet levels; the subband signals include approximate subband signals and detail subband signals; The similar subband signals and detail subband signals of different wavelet levels are screened by using the Pearson correlation coefficient to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal.

4. The method according to claim 3, characterized in that: The step of constructing a feature matrix according to the respiratory effort signal and the wavelet level subband signal comprises: Performing feature extraction on the respiratory effort signal to extract variance features, absolute mean features, root mean square features, respiratory power features, and respiratory frequency features; Comparing the approximate subband signal and the detail subband signal of the wavelet level subband signal, extracting the relative characteristics of the Pearson coefficient, the relative characteristics of the average Euclidean distance and the relative characteristics of the subband variance ratio between the approximate subband signal and the detail subband signal; The standardized signal segments are divided according to the extracted features and relative features to obtain the feature matrix.

5. The method according to claim 4, characterized in that Inputting the feature matrix into the detection model to obtain the number of sleep apnea times includes: Inputting the feature matrix into the trained detection model to obtain the detection results of each standardized signal segment; The number of sleep apnea events of the target population is obtained according to the detection results of the various standardized signal segments.

6. The method according to any one of claims 1 to 5, characterized in that: The detection model is obtained by training a preset extreme gradient boosting machine learning classification model.

7. A sleep apnea detection device, characterized in that: The device comprises: An acquisition module, used for acquiring a target vital sign signal; A reconstruction module, used for performing wavelet transformation on the target vital sign signal to obtain a respiratory effort signal and a wavelet level subband signal corresponding to the respiratory effort signal; An extraction module constructs a feature matrix according to the respiratory effort signal and the wavelet hierarchical subband signal; the feature matrix includes signal features of several dimensions; the signal features are obtained by extracting features from the respiratory effort signal or extracting relative features between subband signals from the wavelet hierarchical subband signal; The detection module is used to input the feature matrix into the detection model to obtain the number of sleep apnea; the detection model is trained using vital sign signals as samples and signal features as annotations.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the sleep apnea detection method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the sleep apnea detection method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the sleep apnea detection method according to any one of claims 1 to 6.

Images