Automatic detection device for sleep apnea syndrome

Abstract

The invention belongs to the technical field of medical monitoring, and particularly discloses a sleep apnea syndrome automatic detection device which comprises a signal acquisition module, a data preprocessing module, a feature extraction module and a fusion decision module. The signal acquisition module is used for acquiring an RR interval signal and a respiratory wave signal; the data preprocessing module is used for removing interference areas through preprocessing and obtaining preprocessed RR interval signals and breath wave signals; the feature extraction module is used for determining sleep apnea events represented by the RR interval signals and the respiratory wave signals through statistical analysis on the basis of the preprocessed RR interval signals and the respiratory wave signals; and the fusion decision module is used for analyzing relevance between sleep apnea events represented by the RR interval signals and the respiratory wave signals and determining a sleep apnea detection result. According to the application, the sleep apnea syndrome can be detected stably and efficiently under the condition that the detection accuracy is guaranteed.

Description

Technical Field

[0001] This application belongs to the technical field of medical monitoring, and more specifically, relates to an automatic detection device for sleep apnea syndrome. Background Art

[0002] Obstructive Sleep Apnea Syndrome (OSAS) is a common sleep disorder, characterized by repeated upper airway collapse during sleep, resulting in airflow obstruction, and further causing hypoventilation or even complete apnea. OSAS not only affects the sleep quality of patients, but also may increase the risk of other diseases.

[0003] In clinical diagnosis, OSAS is usually diagnosed through polysomnography. Polysomnography (PSG) can record physiological signals such as electroencephalogram, eye movement, electromyogram, electrocardiogram, respiratory flow, and blood oxygen saturation, and comprehensively evaluate the severity of OSAS. However, PSG devices are expensive, the detection process is complex, and patients need to be hospitalized for monitoring, which is not suitable for large-scale screening. Therefore, how to find a convenient, efficient, and low-cost screening method and recommend patients with moderate or severe suspected OSAS to the respiratory department for PSG diagnosis has become the focus of current research.

[0004] Currently, the primary screening detection schemes for OSAS mainly include the following: (1) portable sleep monitoring devices, (2) devices based on respiratory rate monitoring, and (3) devices based on dynamic electrocardiogram recorders with dual parameters of electrocardiogram + respiratory wave.

[0005] (1) Portable sleep monitoring devices; Such devices usually monitor parameters such as blood oxygen saturation, respiratory flow, and heart rate. They are more convenient than PSG, and some devices can be used in a home environment. However, due to limited monitoring signals, the detection accuracy is relatively low, it is easily affected by individual physiological differences and environmental factors, there is a high false positive rate, and it is difficult to achieve large-scale screening.

[0006] (2) Devices based on respiratory rate monitoring; This method uses signals such as thoracic impedance and airflow to detect apnea events. However, this type of method is easily affected by thoracic state and body position changes, resulting in certain limitations in detection accuracy, especially the ability to identify transient apnea is weak.

[0007] (3) Devices based on dynamic electrocardiogram recorders with dual parameters of electrocardiogram + respiratory wave; This type of device calculates the change in RR interval through ECG signals, and records respiratory wave data in combination with chest impedance signals. The use of dual-parameter comprehensive analysis can improve the detection accuracy of sleep apnea. Compared with single-parameter detection, this method is a non-invasive screening, and dynamic ECG recording has been widely used in cardiovascular disease detection, so it is suitable for large-scale screening. However, the final diagnosis of this method currently still relies on manual labeling, that is, doctors need to manually mark the starting and ending points of sleep apnea events based on the heart rate grid diagram and respiratory waves, and then determine whether the suspected patient is mild, moderate or severe based on the proportion of attack time.

[0008] The electrocardiogram (ECG) is the manifestation of the heart's electrical activity on the body surface, which reflects the excitement state and excitement propagation process of various parts of the heart. The R wave is the first positive wave in the QRS complex, representing the potential change of ventricular myocyte depolarization (excitation), marking the beginning of ventricular contraction. The RR interval refers to the time interval between two adjacent R waves, reflecting a complete cardiac cycle (the time from ventricular contraction to the next contraction).

[0009] At present, the main method for initial screening of OSAS patients is to use a dynamic electrocardiogram recorder based on dual parameters of electrocardiogram and respiratory wave, but the existing manual labeling method has the following defects.

[0010] (1) It is time-consuming and labor-intensive, and difficult to promote. Doctors need to manually analyze the dynamic electrocardiogram and respiratory wave data, and the labeling time for each patient is 5-10 minutes, which seriously affects the screening efficiency and is difficult to promote to large-scale application.

[0011] (2) Large subjective errors and low labeling consistency. Manual labeling relies on the doctor’s experience, which may vary between different doctors and is easily affected by interference areas in the signal, resulting in low stability of the screening results.

[0012] (3) Difficult to detect in real time. Manual screening usually requires a later data review and analysis, and cannot achieve real-time monitoring, which limits the application of this method in telemedicine and long-term monitoring.

[0013] The manual labeling method is inefficient and the result stability is low. How to stably and efficiently detect sleep apnea syndrome while ensuring the detection accuracy is a technical problem that needs to be solved in this field. Summary of the invention

[0014] In view of the defects of the prior art, the purpose of the present application is to achieve stable and efficient detection of sleep apnea syndrome while ensuring the detection accuracy.

[0015] To achieve the above-mentioned purpose, in a first aspect, the present application provides a sleep apnea syndrome automatic detection device, the device comprising: a signal acquisition module, a data preprocessing module, a feature extraction module and a fusion decision module; A signal acquisition module, configured to acquire RR interval signals and respiratory wave signals; A data preprocessing module, configured to remove interference regions in the RR interval signals and respiratory wave signals through preprocessing, and obtain the preprocessed RR interval signals and respiratory wave signals; A feature extraction module, configured to determine sleep apnea events characterized by the RR interval signals and sleep apnea events characterized by the respiratory wave signals based on the preprocessed RR interval signals and respiratory wave signals through statistical analysis; A fusion decision module, configured to analyze the correlation between the sleep apnea events characterized by the RR interval signals and the sleep apnea events characterized by the respiratory wave signals, and determine the sleep apnea detection result.

[0016] It can be understood that by preprocessing to remove the interference regions in the RR interval signals and respiratory wave signals, it is possible to automatically eliminate interference segments such as arrhythmias (such as atrial fibrillation, premature beats) and motion artifacts, and achieve interference elimination, significantly improving the stability of detection; by automatically acquiring the RR interval signals and respiratory wave signals through the signal acquisition module, automatically performing preprocessing through the data preprocessing module, automatically identifying sleep apnea events through the feature extraction module (i.e., automatically performing behavior feature recognition related to sleep apnea), and automatically generating sleep apnea detection results through the fusion decision module, it is possible to achieve full automation from signal acquisition to result generation, without manual annotation, significantly improving the detection efficiency; by analyzing the correlation between the sleep apnea events characterized by the RR interval signals and the sleep apnea events characterized by the respiratory wave signals, it is possible to effectively fuse the heart rate raster diagram and respiratory wave data, reducing the false positive rate compared to single physiological signal detection methods and ensuring the detection accuracy.

[0017] Therefore, through the cooperation among the data preprocessing module, the feature extraction module, and the fusion decision module, it is possible to stably and efficiently detect sleep apnea syndrome while ensuring the detection accuracy.

[0018] In a possible implementation, the data preprocessing module includes: an RR interval signal preprocessing sub-module and a respiratory wave signal preprocessing sub-module; The RR interval signal preprocessing sub-module is configured to: Identify the motion artifact interference regions in the RR interval signals through a deep learning model; Identify abnormal RR intervals in the RR interval signals through statistical analysis; Remove the motion artifact interference regions and abnormal RR intervals in the RR interval signals; The respiratory wave signal preprocessing sub-module is configured to: Identify the motion artifact interference regions in the respiratory wave signals through a deep learning model; Remove the motion artifact interference region in the respiratory wave signal.

[0019] In a possible implementation, the abnormal RR interval is the abnormal RR interval caused by atrial fibrillation, the abnormal RR interval caused by premature beats, the abnormal RR interval caused by bigeminy, or the abnormal RR interval caused by trigeminy.

[0020] In a possible implementation, the feature extraction module includes: an RR interval signal feature extraction sub-module and a respiratory wave signal feature extraction sub-module; The RR interval signal feature extraction sub-module is used to remove high-frequency noise in the (RR interval) signal, retain the sleep apnea features, and determine the sleep apnea event characterized by the RR interval signal based on the sleep apnea features in the RR interval signal; The respiratory wave signal feature extraction sub-module is used to remove high-frequency noise in the (respiratory wave) signal, retain the sleep apnea features, and determine the sleep apnea event characterized by the respiratory wave signal based on the sleep apnea features in the respiratory wave signal.

[0021] In a possible implementation, the above-mentioned removing high-frequency noise in the signal and retaining the sleep apnea features includes: Perform 5-level decomposition on the signal using db4 wavelet transform to obtain D1, D2, D3, D4, D5, and A5. D1 is the detail coefficient obtained from the first-level decomposition (detail coefficients), D2 is the detail coefficient obtained from the second-level decomposition, D3 is the detail coefficient obtained from the third-level decomposition, D4 is the detail coefficient obtained from the fourth-level decomposition, D5 is the detail coefficient obtained from the fifth-level decomposition, and A5 is the approximation coefficient obtained from the fifth-level decomposition (approximation coefficients); Based on D3, D4, and D5, perform signal reconstruction to retain the sleep apnea features.

[0022] In a possible implementation, the above-mentioned determining the sleep apnea event characterized by the RR interval signal based on the sleep apnea features in the RR interval signal includes: Determine the peaks and valleys in the RR interval signal; Based on the peaks and valleys in the RR interval signal, determine the peak-to-peak distance between two adjacent peaks and the valley-to-valley distance between two adjacent valleys in the RR interval signal; Based on the coefficient of variation threshold, the coefficient of variation of the peak-to-peak distance, and the coefficient of variation of the valley-to-valley distance, determine the start and end points of the sleep apnea event.

[0023] In a possible implementation, determining the start and end points of a sleep apnea event based on the coefficient of variation threshold, the coefficient of variation of the peak-to-peak interval, and the coefficient of variation of the trough-to-trough interval includes: For three adjacent peaks, if the coefficient of variation of the peak-to-peak interval continuously exceeds the coefficient of variation threshold (e.g., 0.3) and the duration exceeds a preset time threshold (e.g., 60 seconds), then mark the start point of the sleep apnea event; or, for three adjacent troughs, if the coefficient of variation of the trough-to-trough interval continuously exceeds the coefficient of variation threshold (e.g., 0.3) and the duration exceeds the time threshold (e.g., 60 seconds), then mark the start point of the sleep apnea event; After marking the start point of the sleep apnea event, if the coefficients of variation of the peak-to-peak interval and the trough-to-trough interval continuously fall below the baseline value and the duration exceeds the time threshold (e.g., 60 seconds), then mark the end point of the sleep apnea event, where the baseline value is less than the coefficient of variation threshold.

[0024] It can be understood that experiments show that when the coefficient of variation of the peak-to-peak interval or the trough-to-trough interval continuously exceeds a preset coefficient of variation threshold (such as 0.3) and this state lasts for more than a preset time threshold (such as 60 seconds), it indicates the start of a sleep apnea event; and after determining the start point, continue to monitor the coefficients of variation of the peak-to-peak interval and the trough-to-trough interval. When these coefficients of variation continuously fall below a set baseline value (the baseline value is less than the coefficient of variation threshold) and the duration exceeds the time threshold, it indicates the end of a sleep apnea event.

[0025] In a possible implementation, determining the sleep apnea event represented by the respiratory wave signal based on the sleep apnea characteristics in the respiratory wave signal includes: Use a sliding window to calculate the change rate of the respiratory wave amplitude; Based on the change rate of the respiratory wave amplitude and the baseline amplitude, determine the start and end points of the sleep apnea event.

[0026] In a possible implementation, determining the start and end points of the sleep apnea event based on the change rate of the respiratory wave amplitude and the baseline amplitude includes: If the change rate of the respiratory wave amplitude continuously falls below the first preset percentage (e.g., 50%) of the baseline amplitude and the duration exceeds the time threshold (e.g., 60 seconds), then mark the start point of the sleep apnea event; After marking the start point of the sleep apnea event, if the change rate of the respiratory wave amplitude continuously exceeds the second preset percentage (e.g., 80%) of the baseline amplitude and the duration exceeds the time threshold (e.g., 60 seconds), then mark the end point of the sleep apnea event, where the first preset percentage is less than the second preset percentage.

[0027] It is understandable that two preset percentages are set. One is used to identify a decrease in the respiratory wave amplitude change rate (such as 50%), and the other is used to identify an increase in the respiratory wave amplitude change rate (such as 80%). And these two percentages satisfy that the first preset percentage is less than the second preset percentage. Experiments show that if the respiratory wave amplitude change rate continuously drops below the first preset percentage of the baseline amplitude, and the duration of this low change rate state exceeds a preset time threshold (such as 60 seconds), it marks the start of a sleep apnea event; once the starting point is calibrated, continue to monitor the amplitude change rate. When it continuously exceeds the second preset percentage of the baseline amplitude, and the duration of this high change rate state exceeds the time threshold, it marks the termination of a sleep apnea event.

[0028] In a possible implementation manner, the sleep apnea detection result includes a first determination result, a second determination result, and a third determination result. The fusion decision module is specifically configured to: If, within the same target time period, the number of occurrences of the sleep apnea event characterized by the RR interval signal is greater than a first preset number of events (for example, 3 times), and the number of occurrences of the sleep apnea event characterized by the respiratory wave signal is greater than the first preset number of events, then generate a first determination result for indicating that the sleep apnea event within the target time period is of high confidence; If, within the same target time period, the number of occurrences of the sleep apnea event characterized by the RR interval signal is greater than the first preset number of events, or the number of occurrences of the sleep apnea event characterized by the respiratory wave signal is greater than the first preset number of events, and within a preset duration before the target time period, the number of occurrences of the sleep apnea event is greater than a second preset number of events (for example, 2 times), and within a preset duration after the target time period, the number of occurrences of the sleep apnea event is greater than the second preset number of events, then generate a second determination result for indicating that the sleep apnea event within the target time period is of high confidence, and the first preset number of events is greater than the second preset number of events; Determine the events other than the sleep apnea events of high confidence as of low confidence, and generate a third determination result for indicating the sleep apnea events of low confidence.

[0029] It is understandable that experiments show that for the RR interval signal and the respiratory wave signal within a specific target time period, if the number of occurrences of the sleep apnea event detected by both of them is greater than a certain number of events (such as 3 times), it marks that the sleep apnea event within this time period is of high confidence; if the number of occurrences of the sleep apnea event detected by any one of the signals is greater than the first preset number of events, and within the preset durations before and after the target time period, the number of occurrences of the sleep apnea event is greater than the second preset number of events (such as 2 times), it also marks that the sleep apnea event within this time period is of high confidence.

[0030] By means of this multi-signal fusion and time period comparison, the accuracy and reliability of sleep apnea event detection can be significantly improved.

[0031] Generally speaking, compared with the prior art, the above technical solutions conceived in this application have the following beneficial effects: (1) Deep learning technology and statistical analysis are used to calibrate the interference regions of RR interval and respiratory wave data, automatically removing interference segments such as arrhythmia (such as atrial fibrillation, premature beats) and motion artifacts, so as to eliminate interference and improve the stability of analysis.

[0032] (2) Automatically calibrate the onset regions of sleep apnea without manual annotation, improving the detection efficiency.

[0033] (3) By using the fusion analysis of RR interval signal and respiratory wave signal, the detection accuracy is improved, and compared with the single physiological signal detection method, the false judgment rate is reduced.

[0034] (4) The computational cost is small. In addition to being suitable for large-scale screening of OSAS by dynamic electrocardiogram, it is also applicable to real-time monitoring and telemedicine, and is suitable for long-term monitoring devices. Description of the Drawings

[0035] Figure 1 It is a schematic structural diagram of an automatic detection device for sleep apnea syndrome provided by an embodiment of this application. Detailed Embodiments

[0036] In order to make the objectives, technical solutions and advantages of this application clearer, the following further describes this application in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application, and are not used to limit this application.

[0037] The terms "first" and "second" etc. in the description and claims of this application are used to distinguish different objects, rather than to describe a specific order of the objects. For example, the first preset percentage and the second preset percentage etc. are used to distinguish different preset percentages, rather than to describe a specific order of the preset percentages.

[0038] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0039] In the description of the embodiments of the present application, unless otherwise specified, "a plurality of" means two or more. For example, a plurality of processing units means two or more processing units, etc.; a plurality of elements means two or more elements, etc.

[0040] The embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application.

[0041] Figure 1 It is a schematic structural diagram of an automatic detection device for sleep apnea syndrome provided by the embodiments of the present application. As Figure 1 shown, the device includes: a signal acquisition module 10, a data preprocessing module 20, a feature extraction module 30, and a fusion decision module 40.

[0042] The signal acquisition module 10: acquires dynamic electrocardiogram signals (specifically RR interval signals) and respiratory wave signals for analyzing sleep apnea events.

[0043] The data preprocessing module 20: performs preprocessing such as denoising, interference area calibration and removal, and standardization on the acquired original signals to improve data quality.

[0044] The feature extraction module 30: extracts key features from the preprocessed signals, such as RR interval fluctuations, respiratory amplitude decrease / recovery trends, etc. Adopts a time series pattern matching algorithm to identify periodic changes in the RR interval raster plot (RR interval signal) and the respiratory wave plot (respiratory wave signal).

[0045] The fusion decision module 40: uses the fusion analysis of the RR interval signal and the respiratory wave signal to determine high-confidence sleep apnea events.

[0046] (1) The signal acquisition module 10 will be exemplarily described below.

[0047] The signal acquisition module can acquire electrocardiogram (ECG, Electrocardiogram) signals through a dynamic electrocardiogram recorder, calculate the RR interval; record respiratory waves through thoracic impedance measurement; synchronously record time stamps to ensure time alignment between different signals and improve the accuracy of data analysis by using statistical methods to eliminate abnormal RR intervals (such as atrial fibrillation, premature beats, bigeminy, trigeminy, etc.).

[0048] (2) The data preprocessing module 20 will be exemplarily described below. The data preprocessing module 20 includes: an RR interval signal preprocessing sub-module and a respiratory wave signal preprocessing sub-module.

[0049] (2-1) The RR interval signal preprocessing sub-module: preprocesses the RR interval signal using deep learning techniques and statistical methods.

[0050] Specifically, the interference regions are calibrated through CNN deep learning to exclude the interference segments caused by movements such as turning over. And statistical methods are used to eliminate abnormal RR intervals such as atrial fibrillation, premature beats, bigeminy, and trigeminy.

[0051] In ambulatory electrocardiogram (ECG), movements such as turning over are interferences not related to heart rhythm. Such interferences can cause signal distortion, making the extraction of RR intervals inaccurate and thus affecting subsequent analysis. Therefore, it is necessary to calibrate and exclude such movement artifact interferences.

[0052] Even if the interference regions of movement artifacts are calibrated, there may still be abnormal intervals in the RR intervals due to arrhythmias (such as atrial fibrillation, premature beats, bigeminy, and trigeminy). Therefore, it is necessary to eliminate the abnormal intervals.

[0053] Exemplarily, a specific way to exclude the interference segments of movement artifacts can be to input the RR interval signal (a time series segment with a length of 256 points, which is one-dimensional data) into a pre-trained CNN model.

[0054] The CNN architecture includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. Input layer: Accepts the standardized RR interval signal. Convolutional layer: Extracts local features (such as waveform morphology) of the RR interval signal. Pooling layer: Reduces the feature dimension and enhances the generalization ability of the model. Fully connected layer: Classifies the signal segment as "interference" or "non-interference".

[0055] The CNN model outputs the calibration result: The positions of the interference segments output by the model will be marked and recorded, and processed for exclusion.

[0056] Exemplarily, a specific way to eliminate abnormal RR intervals can be to eliminate the outliers exceeding 3 times the standard deviation (SD); eliminate the intervals with large fluctuations and poor stability based on the coefficient of variation (CV); and eliminate abnormal RR intervals such as atrial fibrillation, premature beats, bigeminy, and trigeminy through time series pattern analysis.

[0057] Among them, the coefficient of variation is the ratio of the standard deviation to the mean, usually expressed as a percentage.

[0058] Atrial fibrillation: The distribution of RR intervals is highly irregular (CV > 0.3). In atrial fibrillation, the RR intervals are usually irregular, and the abnormal intervals caused by atrial fibrillation can be eliminated by analyzing the coefficient of variation.

[0059] Premature beats: A short interval is followed by a long interval, presenting a typical compensatory pause pattern. Premature beats usually show a suddenly shortened RR interval (i.e., the coupling interval), followed by a longer compensatory interval (lengthened RR interval), and this compensatory pause pattern can be used to identify and eliminate the abnormal intervals caused by atrial fibrillation.

[0060] Bigeminy and trigeminy: The RR intervals show a periodic and fixed pattern. The RR interval characteristics of bigeminy have a fixed alternating pattern: it is manifested as a normal RR interval followed by a premature beat, forming a periodic pattern of "normal - premature beat - normal - premature beat", and this pattern can be used to identify and eliminate the abnormal intervals caused by bigeminy. The RR interval characteristics of trigeminy also have a fixed alternating pattern: usually, it is manifested as two normal RR intervals followed by a premature beat (such as "normal - normal - premature beat"), repeating in cycles, and this pattern can be used to identify and eliminate the abnormal intervals caused by trigeminy.

[0061] Processing results of eliminating abnormal RR intervals: All abnormal intervals are eliminated, and the data in the interference area is filled by interpolation processing, only retaining the RR intervals that truly reflect the regulation of the autonomic nervous system.

[0062] (2-2) Respiratory wave signal preprocessing sub-module: Use deep learning technology to preprocess the respiratory wave signal.

[0063] Specifically, the interference area is identified and excluded through a CNN model to ensure the accuracy of the data. The respiratory wave signal is collected from the body surface impedance and is easily interfered by motion artifacts such as turning over. The interference characteristics include sudden changes in signal amplitude or flat line segments.

[0064] Exemplarily, the respiratory wave signal (a time series segment with a length of 512 points) is input into a pre-trained CNN model.

[0065] The CNN architecture includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. Input layer: Accept the standardized respiratory wave signal. Convolutional layer: Extract local interference characteristics (such as sudden amplitude changes). Pooling layer: Reduce the feature dimension and computational complexity. Fully connected layer: Classify and output the interference area.

[0066] The CNN model outputs a calibration result: Mark the interference area to ensure the signal quality for subsequent analysis.

[0067] (3) Next, an exemplary description of the feature extraction module 30 is given. The feature extraction module 30 includes: an RR interval signal feature extraction sub-module and a respiratory wave signal feature extraction sub-module.

[0068] (3-1) The RR interval signal feature extraction sub-module is used to: Detect the periodic changes in heart rate and identify sleep apnea events through the analysis of the RR interval raster plot.

[0069] The processing logic of the RR interval signal feature extraction sub-module includes the following steps: (3-1-1) Signal time-frequency decomposition, (3-1-2) Signal reconstruction, (3-1-3) Peak and valley detection, (3-1-4) Calculate the peak-to-peak distance and valley-to-valley distance, (3-1-5) Determine sleep apnea events.

[0070] (3-1-1) Signal time-frequency decomposition: Perform 5th-order decomposition using db4 wavelet transform.

[0071] ; Among them, A5 (low-frequency component): Represents the long-term trend and reflects the overall heart rate change trend. D5 - D3 (medium-frequency component): Is related to the rhythmic changes of the autonomic nervous system. D2 - D1 (high-frequency component): May contain noise or short-term changes.

[0072] (3-1-2) Signal reconstruction: Select D5 - D3 for signal reconstruction to retain the main sleep apnea characteristics.

[0073] ; Remove high-frequency noise and retain the heart rate change information related to sleep apnea.

[0074] (3-1-3) Peak and trough detection: Use the first derivative + second derivative method to detect the heart rate fluctuation pattern.

[0075] ; Set the threshold for peak detection: Local maximum value, as follows: And ; Set the threshold for trough detection: Local minimum value, as follows: And .

[0076] (3-1-4) Calculate the peak-to-peak interval and trough-to-trough interval: Detect periodic heart rate changes.

[0077] Calculate the interval between consecutive peaks ( ) and the interval between troughs ( ): ; (3-1-5) Determine sleep apnea events; Calculate the coefficient of variation (CV) of the peak-to-peak interval (or trough-to-trough interval): ; Among them, represents the standard deviation of the peak-to-peak interval (or trough-to-trough interval), represents the mean of the peak-to-peak interval (or trough-to-trough interval). Exemplarily, a sliding window can be set (for example, the sliding window size can be set to 3 minutes), and for the peak-to-peak interval (or trough-to-trough interval) within the sliding window, calculate the standard deviation and the mean, and then divide the standard deviation by the mean to obtain the coefficient of variation.

[0078] When three consecutive peaks / troughs are detected and the coefficient of variation (CV) continuously exceeds a set threshold (such as 0.3) for more than 60 seconds, it is marked as the starting point of sleep apnea.

[0079] When the coefficient of variation of the peak-to-peak interval and trough-to-trough interval returns to the baseline value (CV < 0.1) and remains above for more than 60 seconds, it is marked as the ending point of sleep apnea.

[0080] (3 - 2) The respiratory wave signal feature extraction sub-module is used to: analyze the change in respiratory wave amplitude to assist in identifying sleep apnea events.

[0081] The processing logic of the respiratory wave signal feature extraction sub-module includes the following steps: (3 - 2 - 1) signal time-frequency analysis, (3 - 2 - 2) amplitude calculation, (3 - 2 - 3) determination of the starting point of sleep apnea, (3 - 2 - 4) determination of the ending point of sleep apnea.

[0082] (3 - 2 - 1) Signal time-frequency analysis: Perform 5th-order decomposition using db4 wavelet transform, select D5 - D3 for signal reconstruction, and retain the main sleep apnea features.

[0083] (3 - 2 - 2) Amplitude calculation: Use a 32-point sliding window (which can select 32 points) to calculate the amplitude change rate (ACR) of the respiratory wave: .

[0084] (3 - 2 - 3) Determination of the starting point of sleep apnea: When the ACR continuously drops below 50% of the baseline amplitude for more than 60 seconds, mark the starting point of sleep apnea.

[0085] (3 - 2 - 4) Determination of the ending point of sleep apnea: When the ACR rises above 80% of the baseline amplitude and remains above for more than 60 seconds, mark the ending point of sleep apnea.

[0086] (4) The following is an exemplary description of the fusion decision module 40.

[0087] Rule matching method: If both the RR interval raster plot and the respiratory wave meet the sleep apnea characteristics in the same time period, it is determined as a sleep apnea event with high confidence: Condition A: There are three or more consecutive sleep apnea OSAS events in the RR interval raster plot; Condition B: There are also three or more consecutive sleep apnea OSAS events in the respiratory wave; If condition A and condition B are met, it is determined as a high-confidence sleep apnea event.

[0088] In addition, if within a time period, three or more sleep apnea events are detected through one signal (RR interval raster plot or respiratory wave), but less than three sleep apnea events are detected through the other signal (respiratory wave or RR interval raster plot), then a secondary screening is performed: Case 1: Three or more sleep apnea events are detected through the RR interval raster plot but less than three sleep apnea events are detected through the respiratory wave; check the data for the 3 minutes before and after this time period. If there have been two sleep apnea OSAS events (events detected through RR interval raster plot analysis or events detected through respiratory wave signal analysis), it is still determined as a high-confidence sleep apnea event; Case 2: Three or more sleep apnea events are detected through the respiratory wave but less than three sleep apnea events are detected through the RR interval raster plot; check the data for the 3 minutes before and after. If there have been two sleep apnea OSAS events (events detected through RR interval raster plot analysis or events detected through respiratory wave signal analysis), it is still determined as a high-confidence sleep apnea event.

[0089] (5) The deployment method of the sleep apnea syndrome automatic detection device is exemplarily described below.

[0090] Exemplarily, the sleep apnea syndrome automatic detection device can be divided into a terminal device (such as a wearable device), and the signal acquisition module, data preprocessing module, feature extraction module, and fusion decision-making module are deployed on the terminal device.

[0091] Exemplarily, the sleep apnea syndrome automatic detection device can also be two terminal devices (such as a wearable device and a smartphone). The two terminal devices are communicatively connected. The signal acquisition module is deployed on the first terminal device (such as a wearable device), and the data preprocessing module, feature extraction module, and fusion decision-making module are deployed on the second terminal device (such as a smartphone).

[0092] Exemplarily, the sleep apnea syndrome automatic detection device can be divided into a local device (such as a wearable device) and a cloud server. The local device and the cloud server are communicatively connected. The signal acquisition module is deployed on the local device, and the data preprocessing module, feature extraction module, and fusion decision-making module are deployed on the cloud server.

[0093] This application adopts a solution that combines the RR interval of dynamic electrocardiogram signals with respiratory waves to achieve automatic screening and risk grading of sleep apnea syndrome. It can detect sleep apnea events with high credibility. The system can automatically identify four levels of normal, mildly suspicious, moderately suspicious, and severely suspicious based on the duration of sleep apnea events (the longer the duration, the more severe), and combined with doctor review, recommend moderately and severely high-risk patients to the respiratory department for further diagnosis and treatment.

[0094] It can be understood that the various numerical numbers involved in the embodiments of this application are only for convenience of description and are not used to limit the scope of the embodiments of this application.

[0095] Those skilled in the art can easily understand that the above are only the preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. An automatic detection device for sleep apnea syndrome, characterized in that, Including: A signal acquisition module, a data preprocessing module, a feature extraction module, and a fusion decision-making module; The signal acquisition module is used to acquire RR interval signals and respiratory wave signals; The data preprocessing module is used to remove interference regions in the RR interval signals and respiratory wave signals through preprocessing, and obtain the preprocessed RR interval signals and respiratory wave signals; The feature extraction module is used to determine the sleep apnea events characterized by the RR interval signals and the sleep apnea events characterized by the respiratory wave signals based on the preprocessed RR interval signals and respiratory wave signals through statistical analysis; The fusion decision-making module is used to analyze the correlation between the sleep apnea events characterized by the RR interval signals and the sleep apnea events characterized by the respiratory wave signals, and determine the sleep apnea detection result.

2. The automatic detection device for sleep apnea syndrome according to claim 1, wherein The data preprocessing module includes: an RR interval signal preprocessing sub-module and a respiratory wave signal preprocessing sub-module; The RR interval signal preprocessing sub-module is used for: Identifying the motion artifact interference regions in the RR interval signals through a deep learning model; Identifying abnormal RR intervals in the RR interval signals through statistical analysis; Removing the motion artifact interference regions and abnormal RR intervals in the RR interval signals; The respiratory wave signal preprocessing sub-module is used for: Identifying the motion artifact interference regions in the respiratory wave signals through a deep learning model; Removing the motion artifact interference regions in the respiratory wave signals.

3. The automatic detection device for sleep apnea syndrome according to claim 2, wherein The abnormal RR intervals are the abnormal RR intervals caused by atrial fibrillation, premature beats, bigeminy, or trigeminy.

4. The automatic detection device for sleep apnea syndrome according to claim 1, characterized in that, The feature extraction module includes: an RR interval signal feature extraction sub-module and a respiratory wave signal feature extraction sub-module; The RR interval signal feature extraction sub-module is used to remove high-frequency noise in the signal, retain the sleep apnea features, and determine the sleep apnea events characterized by the RR interval signals based on the sleep apnea features in the RR interval signals; The respiratory wave signal feature extraction sub-module is used to remove high-frequency noise in the signal, retain the sleep apnea features, and determine the sleep apnea events characterized by the respiratory wave signals based on the sleep apnea features in the respiratory wave signals.

5. The automatic detection device for sleep apnea syndrome according to claim 4, wherein The removing high-frequency noise in the signal and retaining the sleep apnea features includes: Performing 5th-order decomposition on the signal using db4 wavelet transform to obtain D1, D2, D3, D4, D5, and A5, where D1 is the detail coefficient obtained from the 1st layer of decomposition, D2 is the detail coefficient obtained from the 2nd layer of decomposition, D3 is the detail coefficient obtained from the 3rd layer of decomposition, D4 is the detail coefficient obtained from the 4th layer of decomposition, D5 is the detail coefficient obtained from the 5th layer of decomposition, and A5 is the approximation coefficient obtained from the 5th layer of decomposition; Based on D3, D4, and D5, performing signal reconstruction to retain the sleep apnea features.

6. The automatic detection device for sleep apnea syndrome according to claim 4, wherein The determining the sleep apnea events characterized by the RR interval signals based on the sleep apnea features in the RR interval signals includes: Determining the peaks and valleys in the RR interval signals; Based on the peaks and troughs in the RR interval signal, determine the peak-to-peak distance between two adjacent peaks and the trough-to-trough distance between two adjacent troughs in the RR interval signal; Based on the coefficient of variation threshold, the coefficient of variation of the peak-to-peak distance, and the coefficient of variation of the trough-to-trough distance, determine the start and end points of the sleep apnea event.

7. The automatic detection device for sleep apnea syndrome according to claim 6, wherein The determining the start and end points of the sleep apnea event based on the coefficient of variation threshold, the coefficient of variation of the peak-to-peak distance, and the coefficient of variation of the trough-to-trough distance includes: For three adjacent peaks, if the coefficient of variation of the peak-to-peak distance continuously exceeds the coefficient of variation threshold and the duration exceeds the time threshold, then mark the start point of the sleep apnea event; or, for three adjacent troughs, if the coefficient of variation of the trough-to-trough distance continuously exceeds the coefficient of variation threshold and the duration exceeds the time threshold, then mark the start point of the sleep apnea event; After marking the start point of the sleep apnea event, if the coefficient of variation of the peak-to-peak distance and the coefficient of variation of the trough-to-trough distance continuously fall below the baseline value and the duration exceeds the time threshold, then mark the end point of the sleep apnea event, where the baseline value is less than the coefficient of variation threshold.

8. The automatic detection device for sleep apnea syndrome according to claim 4, wherein The determining the sleep apnea event characterized by the respiratory wave signal based on the sleep apnea characteristics in the respiratory wave signal includes: Use a sliding window to calculate the change rate of the respiratory wave amplitude; Based on the change rate of the respiratory wave amplitude and the baseline amplitude, determine the start and end points of the sleep apnea event.

9. The automatic detection device for sleep apnea syndrome according to claim 8, wherein The determining the start and end points of the sleep apnea event based on the change rate of the respiratory wave amplitude and the baseline amplitude includes: If the change rate of the respiratory wave amplitude continuously falls below the first preset percentage of the baseline amplitude and the duration exceeds the time threshold, then mark the start point of the sleep apnea event; After marking the start point of the sleep apnea event, if the change rate of the respiratory wave amplitude continuously exceeds the second preset percentage of the baseline amplitude and the duration exceeds the time threshold, then mark the end point of the sleep apnea event, where the first preset percentage is less than the second preset percentage.

10. The automatic detection device for sleep apnea syndrome according to claim 1, characterized in that, The sleep apnea detection result includes a first determination result, a second determination result, and a third determination result. The fusion decision module is specifically configured to: If within the same target time period, the occurrence times of the sleep apnea event characterized by the RR interval signal are greater than the first preset event times, and the occurrence times of the sleep apnea event characterized by the respiratory wave signal are greater than the first preset event times, then generate a first determination result for indicating that the sleep apnea event within the target time period is of high confidence; If within the same target time period, the occurrence times of the sleep apnea event characterized by the RR interval signal are greater than the first preset event times, or the occurrence times of the sleep apnea event characterized by the respiratory wave signal are greater than the first preset event times, and within the preset duration before the target time period, the occurrence times of the sleep apnea event are greater than the second preset event times, and within the preset duration after the target time period, the occurrence times of the sleep apnea event are greater than the second preset event times, then generate a second determination result for indicating that the sleep apnea event within the target time period is of high confidence, where the first preset event times are greater than the second preset event times; Events other than high-confidence sleep apnea events are determined to be of low confidence, and a third determination result is generated to indicate low-confidence sleep apnea events.

Images