Obstructive sleep apnea detection method and device

Abstract

The present invention provides an obstructive sleep apnea detection method and device, which belongs to the field of signal processing technology. The obstructive sleep apnea detection method includes: obtaining physiological data of a person under test and a set of sleep monitoring signals in a preset time period; the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal, and a pulse wave signal; slicing and filtering the sleep monitoring signal to obtain at least one frame of slice signal; inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the person under test; wherein the obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multimodal data detection OSA model.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to an obstructive sleep apnea detection method and device. Background Art

[0002] Obstructive sleep apnea hypopnea syndrome (OSAHS), referred to as OSA, is one of the common sleep breathing disorders, mainly caused by repeated collapse of the upper airway during sleep, repeated cessation of oral and nasal breathing airflow, accompanied by cortical arousal or decreased blood oxygen saturation. This leads to repeated intermittent hypoxia and hypercapnia, with snoring at night, daytime sleepiness, dry mouth in the morning, and headache as the main clinical manifestations. Repeated partial or complete collapse of the upper airway during sleep leads to a continuous reduction (hypopnea) or lack of airflow (apnea) for at least 10 seconds, accompanied by cortical arousal or decreased blood oxygen saturation. OSA is the main cause of excessive sleepiness, reduced quality of life, impaired work performance, and increased risk of motor vehicle collisions. OSA is associated with an increased incidence of hypertension, type 2 diabetes, atrial fibrillation, heart failure, coronary heart disease, stroke, and death. The presence and severity of OSA are usually quantified by the Apnea Hypopnea Index (AHI), which is defined as the number of apneas plus hypopneas per hour of sleep (or number of hours recorded by home testing). OSA is defined as more than 30 recurrent episodes of apnea and hypopnea during 7 hours of sleep per night, or an Apnea Hypopnea Index (AHI) ≥ 5 times / hour.

[0003] There are two existing methods for detecting OSA. One is to use PSG equipment for polysomnography. This method requires a sleep technician to be present during the test, and the test must be carried out in a sleep laboratory all night, which consumes a lot of manpower. The test results are generally obtained manually by sleep technicians based on their experience, which makes the test inefficient. In addition, the technician only considers the person being tested to have OSA when obstructive sleep apnea events cause obvious symptoms, and sleep apnea without any symptoms in the sleep record is usually not considered to be OSA, which makes the OSA test results inaccurate. The other is to use deep learning methods for detection, but most of the existing studies only use electrocardiogram signals, which makes the classification accuracy of respiratory events low, and there is no processing of data that has not entered the sleep state. Summary of the invention

[0004] The present invention provides an obstructive sleep apnea detection method and device, which are used to solve the defect that the OSA detection result in the prior art is not accurate enough and improve the accuracy of the OSA detection result.

[0005] The present invention provides an obstructive sleep apnea detection method, comprising: obtaining physiological data of a person under test and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal comprises at least one of an electrocardiogram signal, a blood oxygen signal and a pulse wave signal; performing slicing processing and filtering processing on the sleep monitoring signal to obtain at least one frame of sliced ​​signals; inputting the physiological data and the at least one frame of sliced ​​signals into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the person under test; wherein the obstructive sleep apnea detection model comprises a sleep stage classification model, a respiratory event classification model, a high-level sample classification model and a multimodal data detection OSA model.

[0006] According to an obstructive sleep apnea detection method provided by the present invention, the physiological data and the at least one frame of slice signal are input into a successfully trained obstructive sleep apnea detection model to obtain the severity of obstructive sleep apnea hypopnea syndrome (OSA) of the tested person, including: inputting the at least one frame of slice signal into a successfully trained sleep stage classification model to obtain a sleep stage high-level feature vector corresponding to each frame of the slice signal; inputting the at least one frame of slice signal into a successfully trained respiratory event classification model to obtain a respiratory event classification high-level feature vector corresponding to each frame of the slice signal; inputting the sleep stage high-level feature vector corresponding to each frame of the slice signal and the respiratory event classification high-level feature vector corresponding to each frame of the slice signal into a successfully trained high-level sample classification model to obtain a feature vector corresponding to each frame of the slice signal; integrating the feature vectors of all the slice signals to calculate the proportion of all slice signals in a sleep state and with respiratory events in all slice signals; inputting the proportion result and the physiological data into a successfully trained multimodal data OSA detection model to obtain the severity of obstructive sleep apnea of ​​the tested person.

[0007] According to an obstructive sleep apnea detection method provided by the present invention, the method also includes: training the obstructive sleep apnea detection model; the training of the obstructive sleep apnea detection model includes: constructing the obstructive sleep apnea detection model; obtaining a group of sleep monitoring signal samples of at least one subject, a sleep stage label corresponding to the subject, a respiratory event label corresponding to the subject, an OSA category label corresponding to the subject, and the physiological data of the subject; slicing and filtering each of the sleep monitoring signal samples of the subject to obtain at least one frame of slice signal samples; according to the sleep stage label corresponding to each of the subjects and the respiratory event label corresponding to the subject, obtaining the sleep stage label and the corresponding respiratory event label corresponding to each frame of the slice signal sample; constructing a training sample set; wherein the training sample set includes at least one training sample, each of the training samples includes all slice signal samples of a subject, the physiological data of the subject, the sleep stage label corresponding to each frame of the slice signal sample, the respiratory event label corresponding to each frame of the slice signal sample, and the OSA category label corresponding to the subject; using the training sample set to train the obstructive sleep apnea detection model.

[0008] According to an obstructive sleep apnea detection method provided by the present invention, the obstructive sleep apnea detection model is trained by using the training sample set, including: training the sleep staging classification model according to all the slice signal samples in the training samples and the sleep staging labels corresponding to the slice signal samples in each frame, and obtaining a successfully trained sleep staging classification model after the training is completed; training the respiratory event classification model according to all the slice signal samples in the training samples and the respiratory event labels corresponding to the slice signal samples in each frame, and obtaining a successfully trained respiratory event classification model after the training is completed; inputting all the slice signal samples in the training samples into the successfully trained sleep staging classification model to obtain sleep staging high-level feature samples corresponding to the slice signal samples in each frame, inputting all the slice signal samples in the training samples into the successfully trained respiratory event classification model to obtain respiratory event classification high-level feature samples corresponding to the slice signal samples in each frame; The sleep staging high-level feature samples and respiratory event classification high-level feature samples corresponding to the slice signal samples, the sleep staging labels corresponding to each frame of the slice signal samples, and the respiratory event labels corresponding to each frame of the slice signal samples are used to train the high-level sample classification model, and a successfully trained high-level sample classification model is obtained after the training is completed; the sleep staging high-level feature samples corresponding to each frame of the slice signal samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples are input into the successfully trained high-level sample classification model to obtain the feature vectors of all the slice signal samples; the feature vectors of all the slice signal samples are integrated to calculate the proportion of all slice signal samples that are in a sleep state and have respiratory events in all slice signal samples; according to the proportion results, the physiological data of the subjects in the training samples and the OSA category labels corresponding to the subjects, the multimodal data OSA detection model is trained, and the successfully trained multimodal data OSA detection model is obtained after the training is completed.

[0009] According to an obstructive sleep apnea detection method provided by the present invention, the training of the sleep stage classification model includes: inputting all the slice signal samples in the training samples into the sleep stage classification model to obtain sleep stage high-level feature samples corresponding to each frame of the slice signal samples; calculating the loss value of the sleep stage classification model according to the sleep stage high-level feature samples corresponding to each frame of the slice signal samples in each of the training samples and the sleep stage labels corresponding to each frame of the slice signal samples, and using the loss value to update the parameters of the sleep stage classification model.

[0010] According to an obstructive sleep apnea detection method provided by the present invention, all the slice signal samples in the training samples are input into the sleep staging classification model to obtain the sleep staging high-level feature samples corresponding to each frame of the slice signal samples, including: inputting all the slice signal samples into the neural network of the sleep staging classification model to extract the high-level representation samples of each frame of the slice signal samples; inputting the high-level representation samples of each frame of the slice signal samples into multiple inverted residual modules with grouping attention mechanisms of the sleep staging classification model, density clustering the high-level representation samples belonging to the same type of OSA samples, and obtaining a high-level representation subset of each type of OSA; inputting the high-level representation subset of each type of OSA into the average pooling layer and the fully connected layer of the sleep staging classification model to obtain the sleep staging high-level feature samples corresponding to each frame of the slice signal samples.

[0011] According to an obstructive sleep apnea detection method provided by the present invention, the training of the respiratory event classification model includes: inputting all the slice signal samples in the training samples into the respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples; calculating the loss value of the respiratory event classification model according to the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples in each of the training samples and the respiratory event labels corresponding to each frame of the slice signal samples, and using the loss value to update the parameters of the respiratory event classification model.

[0012] According to an obstructive sleep apnea detection method provided by the present invention, the training of the high-level sample classification model includes: inputting the sleep staging high-level feature samples and the respiratory event classification high-level feature samples corresponding to the slice signal samples of each frame into the high-level sample classification model to obtain the feature vector of the slice signal samples of each frame; calculating the loss value of the high-level sample classification model according to the feature vector of the slice signal samples of each frame in each of the training samples, the sleep staging label corresponding to the slice signal samples of each frame, and the respiratory event label corresponding to the slice signal samples of each frame, and using the loss value to update the parameters of the high-level sample classification model.

[0013] According to an obstructive sleep apnea detection method provided by the present invention, the training of the multimodal data detection OSA model includes: inputting the proportion result and the physiological data of the subject in the training sample into the multimodal data detection OSA model, obtaining the OSA disease degree of the subject corresponding to the training sample, calculating the loss value of the multimodal data detection OSA model according to the OSA disease degree of the subject corresponding to each training sample and the OSA category label corresponding to the subject, using the loss value to update the parameters of the multimodal data detection OSA model, and obtaining the successfully trained multimodal data detection OSA model after the training is completed.

[0014] The present invention also provides an obstructive sleep apnea detection device, comprising:

[0015] A data acquisition module, used to acquire physiological data of the person being tested and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal, and a pulse wave signal;

[0016] A signal processing module, used for performing slicing and filtering on the sleep monitoring signal to obtain at least one frame of slice signal;

[0017] A result acquisition module is used to input the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the tested person;

[0018] The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model and a multimodal data detection OSA model.

[0019] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the obstructive sleep apnea detection method as described above is implemented.

[0020] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the obstructive sleep apnea detection method as described in any one of the above is implemented.

[0021] The obstructive sleep apnea detection method and device provided by the present invention first obtain the physiological data of the person to be tested and a group of sleep monitoring signals in a preset time period, then slice and filter the sleep monitoring signals to obtain multiple frames of slice signals, and finally input the multiple frames of slice signals into a successfully trained obstructive sleep apnea detection model to automatically predict the OSA severity of the person to be tested, thereby reducing a large amount of manual intervention in the OSA detection process, improving the detection efficiency, and at the same time improving the accuracy. Moreover, since each detection uses the same successfully trained obstructive sleep apnea detection model, the stability of the detection result is also improved to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic flow chart of the obstructive sleep apnea detection method provided by the present invention;

[0024] Figure 2 It is a flow chart of obtaining the degree of OSA of a person under test by using a successfully trained obstructive sleep apnea detection model provided by the present invention;

[0025] Figure 3 It is a schematic diagram of the process of training an obstructive sleep apnea detection model provided by the present invention;

[0026] Figure 4 It is a schematic diagram of the process of processing slice signal samples in training samples by the sleep stage classification model provided by the present invention;

[0027] Figure 5 is a schematic structural diagram of an obstructive sleep apnea detection device provided by the present invention;

[0028] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The terms "first", "second", etc. in the specification and claims of the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0031] It should be understood that the reference to "one embodiment" or "an embodiment" in the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, the "in one embodiment" or "in an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0032] Combine the following Figure 1-Figure 4 The obstructive sleep apnea detection method of the present invention is described.

[0033] Figure 1 A schematic diagram of the process of detecting obstructive sleep apnea provided by the present invention. Figure 1 As shown, the method includes:

[0034] Step 101, obtaining physiological data of a person under test and a set of sleep monitoring signals in a preset time period.

[0035] It should be noted that it is necessary to ensure that the person being measured has entered a sleep state within a preset time period, and the specific duration of the preset time period can be customized according to actual needs, for example, 7 hours.

[0036] It should be noted that the physiological data of the person being measured includes but is not limited to the following: the height, weight, age, gender, BMI and neck circumference of the person being measured.

[0037] In the related art, when detecting whether a person being tested suffers from OSA, usually only a single physiological signal is selected for judgment, such as judging by electrocardiogram or blood oxygen, which may make the detection result inaccurate.

[0038] In the present invention, a set of sleep monitoring signals may include a variety of different physiological signals to improve the accuracy of the final detection result.

[0039] Optionally, the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal and a pulse wave signal.

[0040] In one embodiment, the sleep monitoring signal includes but is not limited to an electroencephalogram signal, an electrooculogram signal, a mandibular electromyography signal, an electrocardiogram signal, a nasal airflow signal, a chest and abdominal movement signal, and a blood oxygen saturation signal.

[0041] Optionally, physiological signals collected by a wearable device or a PSG portable device are obtained.

[0042] It should be noted that wearable devices or PSG portable devices are easier to operate than traditional PSG devices, making the OSA detection process more convenient.

[0043] It should be noted that each physiological signal is a one-dimensional signal.

[0044] It is understandable that the electronic device obtains the physiological data of the person being measured, and at the same time obtains a set of sleep monitoring signals collected by the wearable device or PSG portable device during a preset time period.

[0045] Step 102: Slice and filter the sleep monitoring signal to obtain at least one frame of slice signal.

[0046] It should be noted that the slicing processing method is to cut each physiological signal in the sleep monitoring signal into multiple frames of slice signals according to a preset duration, wherein the preset duration can be customized according to actual needs.

[0047] In one implementation, each physiological signal is cut into multiple frame slice signals at 30 seconds intervals.

[0048] In another embodiment, each physiological signal is cut into multiple frame slice signals according to 60 seconds.

[0049] It can be understood that after the sleep monitoring signal is sliced, it is also necessary to perform filtering processing, the purpose of which is to remove irrelevant noise on the signal, improve the signal-to-noise ratio of the signal, and improve the reliability of the signal.

[0050] The method used in the filtering process may be an existing filtering method, including but not limited to low-pass filtering, FFT filtering or wavelet denoising filtering.

[0051] It can be understood that the electronic device performs slice processing and filtering processing on the sleep monitoring signal according to a preset time length, and then obtains at least one frame of slice signal.

[0052] Step 103: Input the physiological data and the at least one frame of slice signals into the successfully trained obstructive sleep apnea detection model to obtain the degree of the subject's obstructive sleep apnea hypopnea syndrome (OSA).

[0053] The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multi-modal data detection OSA model.

[0054] The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multi-modal data detection OSA model.

[0055] It can be understood that the input of the obstructive sleep apnea detection model is the physiological data of the subject and the slice signals, and the output is the degree of the subject's OSA.

[0056] It should be noted that the sleep stage classification model is used to classify the sleep stage corresponding to the slice signals;

[0057] The respiratory event classification model is used to classify whether a respiratory event occurs in the slice signals;

[0058] In the present invention, a respiratory event refers to the occurrence of apnea or hypopnea.

[0059] The high-level sample classification model is used to classify whether the slice signals are in the sleep state and a respiratory event occurs;

[0060] The multi-modal data detection OSA model is used to predict the degree of the subject's OSA.

[0061] It should be noted that the degree of OSA can be divided by the value of AHI. For example, when AHI < 5, the subject has no OSA condition, that is, normal; when 5 < AHI < 15, the degree of the subject's OSA is mild; when 15 < AHI < 30, the degree of the subject's OSA is moderate; when AHI > 30, the degree of the subject's OSA is severe.

[0062] The obstructive sleep apnea detection method provided by the present invention first obtains the physiological data of the person to be tested and a group of sleep monitoring signals in a preset time period, then slices and filters the sleep monitoring signals to obtain multiple frames of sliced ​​signals, and finally inputs the multiple frames of sliced ​​signals into a successfully trained obstructive sleep apnea detection model to automatically predict the OSA severity of the person to be tested, thereby reducing a large amount of manual intervention in the OSA detection process, improving detection efficiency, and at the same time improving accuracy. Moreover, since each detection uses the same successfully trained obstructive sleep apnea detection model, the stability of the detection result is also improved to a certain extent.

[0063] Optionally, Figure 2 FIG. 1 is a flow chart of obtaining the degree of OSA of a person under test by using a successfully trained obstructive sleep apnea detection model provided by the present invention, such as Figure 2 The step 102 shown includes steps 1021 to 1025:

[0064] Step 1021: input the at least one frame of slice signal into a successfully trained sleep stage classification model to obtain a sleep stage high-level feature vector corresponding to each frame of the slice signal.

[0065] It should be noted that the sleep stage high-level feature vector can be understood as a feature vector corresponding to the slice signal, and the sleep period corresponding to the slice signal and the corresponding probability can also be obtained through the feature vector.

[0066] The sleep period includes the wakefulness period, non-rapid eye movement sleep N period and rapid eye movement sleep REM period. When the sleep period corresponding to the slice signal is N period or REM period, it indicates that the person being tested enters a sleep state in the time frame corresponding to the slice signal.

[0067] It can be understood that the input of the sleep stage classification model is the slice signal, and the output is the sleep stage high-level feature vector corresponding to the slice signal.

[0068] Step 1022: input the at least one frame of slice signal into a successfully trained respiratory event classification model to obtain a respiratory event classification high-level feature vector corresponding to each frame of the slice signal.

[0069] It should be noted that the respiratory event classification high-level feature vector can be understood as a feature vector corresponding to the slice signal, and whether a respiratory event occurs in the slice signal and the probability can also be obtained through the feature vector.

[0070] It can be understood that the input of the respiratory event classification model is the slice signal, and the output is the respiratory event classification high-level feature vector corresponding to the slice signal.

[0071] It should be noted that the sleep stage classification model and the respiratory event classification model are two independent models. The slice signal can be first input into the sleep stage classification model and then into the respiratory event classification model. The slice signal can also be first input into the respiratory event classification model and then into the sleep stage classification model. The slice signal can also be input into the sleep stage classification model and the respiratory event classification model at the same time. The present invention does not limit the order in which the slice signal is input into the sleep stage classification model and the respiratory event classification model.

[0072] Step 1023, input the sleep stage high-level feature vector corresponding to each frame of the slice signal and the respiratory event classification high-level feature vector corresponding to each frame of the slice signal into the successfully trained high-level sample classification model to obtain the feature vector corresponding to each frame of the slice signal.

[0073] It should be noted that the feature vector corresponding to each frame of the slice signal can be understood as the result of simultaneously classifying the sleep period of the slice signal and whether a respiratory event occurs. For example, the feature vector corresponding to each frame of the slice signal can indicate that a slice signal is in period N and a respiratory event occurs, or the feature vector corresponding to each frame of the slice signal can indicate that a slice signal is in the awake period and no respiratory event occurs, and so on.

[0074] It can be understood that the input of the high-level sample classification model includes two parts, one part is the sleep stage high-level feature vector output by the sleep stage classification model, and the other part is the respiratory event classification high-level feature vector output by the respiratory event classification model. The output is the feature vector corresponding to the slice signal.

[0075] Step 1024: Integrate the feature vectors of all the slice signals, and calculate the proportion of all slice signals that are in a sleeping state and have respiratory events in all slice signals.

[0076] It should be noted that the calculation result of the proportion of all slice signals that are in a sleep state and have respiratory events in all slice signals is for one person being tested. The corresponding proportion result of the person being tested is the ratio of the number of frames of all slice signals that are in a sleep state and have respiratory events to the number of frames of all slice signals of the person being tested.

[0077] It should be understood that the input and output of the sleep stage classification model, respiratory event classification model, and high-level sample classification model are all for slice signals. By calculating the proportion of all slice signals that are in a sleep state and have respiratory events in all slice signals, the relevant information describing the slice signals can be converted into information that can describe the AHI value of the person being measured.

[0078] The proportion result of a person to be measured is positively correlated with the AHI value of the person to be measured, but it is not the true AHI value.

[0079] Step 1025: Input the proportion result and the physiological data into the successfully trained multi-modal data detection OSA model to obtain the severity of the obstructive sleep apnea suffered by the person to be measured.

[0080] It should be noted that although the frame number proportion result is not the true AHI value, the frame number proportion result can still be used to reflect the severity of the obstructive sleep apnea suffered by the person to be measured. For example: when the frame number proportion result < 5% (corresponding to AHI < 5), the person to be measured has no OSA condition, that is, normal; when 5% < frame number proportion result < 15% (corresponding to 5 < AHI < 15), the OSA disease degree of the person to be measured is mild; when 15% < frame number proportion result < 30% (corresponding to 15 < AHI < 30), the OSA disease degree of the person to be measured is moderate; when the frame number proportion result > 30% (corresponding to AHI > 30), the OSA disease degree of the person to be measured is severe.

[0081] It can be understood that the input of the multi-modal data detection OSA model is the proportion result of all slice signals corresponding to the person to be measured that are in the sleep state and have respiratory events among all slice signals and the physiological data of the person to be measured, and the output is the severity of the obstructive sleep apnea suffered by the person to be measured.

[0082] In the embodiment of the present invention, first, the slice signals of the person to be measured are used for feature extraction by the successfully trained sleep stage classification model to obtain the corresponding high-level sleep stage feature vectors, and the slice signals of the person to be measured are used for feature extraction by the successfully trained respiratory event classification model to obtain the corresponding high-level respiratory event classification feature vectors. Then, the high-level sleep stage feature samples and high-level respiratory event classification feature samples corresponding to each frame of slice signals are input into the successfully trained high-level sample classification model to obtain the feature vectors corresponding to each frame of slice signals. Next, the proportion result of all slice signals that are in the sleep state and have respiratory event states among all slice signals is calculated. Finally, according to the proportion result, the OSA severity of the person to be measured is predicted by the successfully trained multi-modal data detection OSA model, realizing the automatic detection of OSA. Since both the sleep state and respiratory events of the person to be measured are considered in the detection process, compared with the traditional method of detecting OSA only by referring to respiratory events, the detection result is more accurate, and since the same model is used for each detection, the stability of the detection result is also improved to a certain extent.

[0083] Optionally, Figure 3 is a schematic flow chart of training an obstructive sleep apnea detection model provided by the present invention, as Figure 3 shown:

[0084] The method further comprises: training the obstructive sleep apnea detection model;

[0085] The training of the obstructive sleep apnea detection model comprises:

[0086] Step 301: construct an obstructive sleep apnea detection model.

[0087] It should be noted that the obstructive sleep apnea model is constructed by constructing a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multimodal data detection OSA model.

[0088] Step 302, obtaining a set of sleep monitoring signal samples of at least one subject, a sleep stage label corresponding to the subject, a respiratory event label corresponding to the subject, an OSA category label corresponding to the subject, and physiological data of the subject.

[0089] It should be noted that the sleep stage label corresponding to the subject, the respiratory event label corresponding to the subject, and the OSA category label corresponding to the subject have been determined in advance, and the electronic device can directly obtain these data when they are needed.

[0090] Among them, the determination of the label can be carried out according to the following steps: first, use the PSG device to obtain the subject's polysomnography, including electroencephalogram (C3 / A2, C4 / A1), electrooculogram, electromyogram, electrocardiogram, nasal airflow, chest and abdominal movement and pulse oximetry, and then score the polysomnography according to the standards of the American Academy of Sleep Medicine to obtain the subject's sleep stage label and respiratory event label, and finally analyze the number of apnea and hypopnea in the subject from the polysomnography, and use the definition of the apnea-hypopnea index (AHI), that is, the number of apnea and hypopnea per hour of sleep time, to calculate the subject's AHI value, and then obtain the subject's OSA category label.

[0091] The sleep stage label corresponding to the subject can reflect the sleep period of the subject in each time frame of the sleep monitoring signal sample, that is, the subject is in the awake period, N period or REM period.

[0092] The respiratory event label corresponding to the subject can reflect whether the subject has a respiratory event in each time frame of the sleep monitoring signal sample.

[0093] The OSA category label corresponding to the subject can reflect the severity of OSA in the subject.

[0094] It should be noted that the subject's physiological data includes but is not limited to height, weight, age, gender, BMI and neck circumference.

[0095] The sleep monitoring signals of the subject include, but are not limited to, electroencephalogram signals, electrooculogram signals, mandibular electromyography signals, electrocardiogram signals, nasal airflow signals, chest and abdominal movement signals, and blood oxygen saturation signals.

[0096] Step 303: Slice and filter the sleep monitoring signal samples of each subject to obtain at least one frame of slice signal samples.

[0097] It should be noted that the slicing processing method is to cut each physiological signal in the sleep monitoring signal sample into multiple frames of slice signal samples according to a preset duration, wherein the preset duration can be customized according to actual needs.

[0098] The method used in the filtering process may be an existing filtering method, including but not limited to low-pass filtering, FFT or wavelet denoising filtering and the like.

[0099] Step 304 , obtaining the sleep stage label and the corresponding respiratory event label corresponding to each frame of the slice signal sample according to the sleep stage label corresponding to each subject and the respiratory event label corresponding to the subject.

[0100] It should be noted that the sleep stage label and the corresponding respiratory event label corresponding to the subject are labels corresponding to the person. It can also be understood that the sleep stage label and the respiratory event label corresponding to the subject are labels corresponding to the complete sleep monitoring signal sample. Since the slice signal sample is a partial signal slice frame in the sleep monitoring signal sample, the sleep stage label and the corresponding respiratory event label corresponding to each frame of the slice signal sample can be obtained from the sleep stage label and the respiratory event label corresponding to the subject.

[0101] Step 305: construct a training sample set.

[0102] The training sample set includes at least one training sample, each of which includes all slice signal samples of a subject, physiological data of the subject, sleep stage labels corresponding to each frame of the slice signal samples, respiratory event labels corresponding to each frame of the slice signal samples, and OSA category labels corresponding to the subject.

[0103] Step 306: Use the training sample set to train the obstructive sleep apnea detection model.

[0104] It should be noted that the training sample set can be divided into a training set, a validation set and a test set to train the obstructive sleep apnea detection model. For example, the training sample set is divided into a training set, a validation set and a test set in equal proportions.

[0105] In an embodiment of the present invention, the sleep monitoring data of each subject is sliced ​​and filtered using the existing sleep monitoring data of multiple subjects to obtain multi-frame slice signal sample data corresponding to each subject, and then a training sample set is constructed using the slice signal sample data corresponding to the subject, the sleep stage label corresponding to the subject, the respiratory event label corresponding to the subject, the OSA category label corresponding to the subject, and the physiological data of the subject. The training sample set is used to train an obstructive sleep apnea detection model. The trained obstructive sleep apnea detection model can predict the severity of OSA, thereby reducing excessive manual intervention in the OSA detection process and improving the monitoring efficiency and the accuracy of the detection results.

[0106] Optionally, the using the training sample set to train the obstructive sleep apnea detection model includes:

[0107] Training the sleep stage classification model according to all the slice signal samples in the training sample and the sleep stage labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained sleep stage classification model after the training is completed;

[0108] Training the respiratory event classification model according to all the slice signal samples in the training sample and the respiratory event labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained respiratory event classification model after the training is completed;

[0109] Input all the slice signal samples in the training samples into the successfully trained sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal samples; input all the slice signal samples in the training samples into the successfully trained respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples;

[0110] According to the sleep staging high-level feature samples and respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples, the sleep staging labels corresponding to each frame of the slice signal samples, and the respiratory event labels corresponding to each frame of the slice signal samples, the high-level sample classification model is trained to obtain a successfully trained high-level sample classification model after the training is completed;

[0111] Input the sleep stage high-level feature samples corresponding to each frame of the slice signal samples and the sleep stage high-level feature samples corresponding to each frame of the slice signal samples into the successfully trained high-level sample classification model to obtain feature vectors of all the slice signal samples;

[0112] Integrate the feature vectors of all the slice signal samples, and calculate the proportion of all slice signal samples that are in a sleeping state and have respiratory events in all slice signal samples;

[0113] The multimodal data OSA detection model is trained according to the proportion results, the physiological data of the subject in the training sample, and the OSA category label corresponding to the subject, and the successfully trained multimodal data OSA detection model is obtained after the training is completed.

[0114] It is important to understand that the sleep stage classification model, respiratory event classification model, high-level sample classification model, and multimodal data detection OSA model are equivalent to classifiers;

[0115] The classifier may be an existing classifier, such as xgboost, lightgbm, a neural network, a support vector machine, a decision tree classifier, or a Bayesian decision classifier.

[0116] It should be noted that a successfully trained sleep stage classification model, a successfully trained respiratory event classification model, a successfully trained high-level sample classification model, and a successfully trained multimodal data OSA detection model constitute a successfully trained obstructive sleep apnea detection model.

[0117] It should be noted that there is no requirement for the training order of the sleep stage classification model and the respiratory event classification model. The sleep stage classification model can be trained first and then the respiratory event classification model, or the respiratory event classification model can be trained first and then the sleep stage classification model, or the sleep stage classification model and the respiratory event classification model can be trained at the same time. The high-level sample classification model can only be trained when both the sleep stage classification model and the respiratory event classification model are successfully trained; the multimodal data OSA detection model can only be trained when the high-level sample classification model is successfully trained.

[0118] Optionally, before inputting the training sample set into the obstructive sleep apnea detection model, the generalization ability of the obstructive sleep apnea detection model can be improved by data enhancement to prevent overfitting, wherein the data enhancement method includes but is not limited to: generating adversarial networks, data training weight adjustment, upsampling and downsampling, etc.

[0119] Optionally, before inputting the training sample set into the obstructive sleep apnea detection model, feature engineering is also required for the physiological data of each subject in the training sample, wherein the feature engineering includes but is not limited to: feature normalization, feature fusion, principal component analysis, local preservation projection, Laplace eigenmap, local linear embedding and linear discriminant analysis.

[0120] In an embodiment of the present invention, a sleep staging classification model, a respiratory event classification model, a high-level sample classification model, and a multimodal data OSA detection model are trained one by one through slice signal samples, sleep staging labels, respiratory event labels, OSA category labels, and physiological data of the subjects in the training samples. Then, a successfully trained obstructive sleep apnea detection model is constructed by the successfully trained models, which provides model support for predicting whether the subject suffers from OSA and the severity of OSA.

[0121] Optionally, the training of the sleep stage classification model includes:

[0122] Inputting all the slice signal samples in the training sample into the sleep stage classification model to obtain sleep stage high-level feature samples corresponding to each frame of the slice signal sample;

[0123] According to the sleep stage high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the sleep stage labels corresponding to the slice signal samples in each frame, the loss value of the sleep stage classification model is calculated, and the parameters of the sleep stage classification model are updated using the loss value.

[0124] It should be noted that the sleep stage high-level feature samples output by the sleep stage classification model are equivalent to the predicted values, and the sleep stage labels are equivalent to the actual values. Therefore, the existing loss function is used to calculate the loss value between the predicted value and the actual value, and then the calculated loss value is used to update the parameters of the sleep stage classification model.

[0125] In an embodiment of the present invention, the loss value is calculated using the sleep stage high-level feature samples output by the sleep stage classification model and the sleep stage label of each frame slice signal sample, and then the sleep stage classification model is updated in reverse. Continuous training according to this step can obtain a successfully trained sleep stage classification model, which provides basic model support for obtaining a successfully trained obstructive sleep apnea detection model.

[0126] Optionally, Figure 4 FIG. 1 is a flow chart of the sleep stage classification model provided by the present invention processing slice signal samples in the training samples. Figure 4 As shown, the step of inputting all the slice signal samples in the training sample into the sleep stage classification model of the obstructive sleep apnea detection model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal sample includes:

[0127] Step 401: input all the slice signal samples into the neural network of the sleep stage classification model, and extract high-level representation samples of each frame of the slice signal samples.

[0128] It should be noted that the choice of neural network includes but is not limited to residual neural network, convolutional neural network based on inverted residual module, etc.

[0129] Step 402, input the high-level representation samples of the slice signal samples of each frame into multiple inverted residual modules with grouping attention mechanism of the sleep stage classification model, and perform density clustering on the high-level representation samples belonging to the same type of OSA samples to obtain a high-level representation subset of each type of OSA.

[0130] Step 403: Input the high-level representation subset of each type of OSA into the average pooling layer and the fully connected layer of the sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal sample.

[0131] It can be understood that the structure of the sleep stage classification model includes a neural network, multiple inverted residual modules with grouped attention mechanism, average pooling layer and fully connected layer.

[0132] It should be noted that the structure of the respiratory event classification network in the present invention is completely consistent with the structure of the sleep stage classification network.

[0133] In the embodiment of the present invention, the neural network of the sleep stage classification model is first used to extract the high-level representation samples of the slice signal samples of each frame, and then the multiple inverted residual modules with grouped attention mechanisms of the sleep stage classification model are used to perform density clustering on all high-level representation samples to obtain the high-level representation subsets of each type of OSA, and finally the high-level representation subsets of each type of OSA are input into the average pooling layer and the fully connected layer of the sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal samples. The various modules within the sleep stage classification model cooperate with each other to jointly obtain the sleep stage prediction value of the slice signal sample, which provides partial data for calculating the loss value of the sleep stage classification model.

[0134] Optionally, the training of the respiratory event classification model comprises:

[0135] Inputting all the slice signal samples in the training sample into the respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples;

[0136] According to the respiratory event classification high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the respiratory event labels corresponding to the slice signal samples in each frame, the loss value of the respiratory event classification model is calculated, and the parameters of the respiratory event classification model are updated using the loss value.

[0137] In an embodiment of the present invention, the loss value is calculated using the respiratory event classification high-level feature samples output by the respiratory event classification model and the respiratory event label of each frame slice signal sample, and then the respiratory event classification model is updated in reverse. Continuous training according to this step can obtain a successfully trained respiratory event classification model, which provides basic model support for obtaining a successfully trained obstructive sleep apnea detection model.

[0138] Optionally, the training of the high-level sample classification model includes:

[0139] Input the sleep stage high-level feature samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples into the high-level sample classification model to obtain a feature vector of each frame of the slice signal samples;

[0140] According to the feature vector of the slice signal sample in each frame of each training sample, the sleep stage label corresponding to the slice signal sample in each frame, and the respiratory event label corresponding to the slice signal sample in each frame, the loss value of the high-level sample classification model is calculated, and the parameters of the high-level sample classification model are updated using the loss value.

[0141] In an embodiment of the present invention, the loss value is calculated using the feature vector output by the high-level sample classification model, the sleep stage label of each frame slice signal, and the respiratory event label, and then the high-level sample classification model is updated in reverse. By continuously training according to this step, a successfully trained high-level sample classification model can be obtained, which provides basic model support for obtaining a successfully trained obstructive sleep apnea detection model.

[0142] Optionally, the training of the multimodal data OSA detection model includes:

[0143] The proportion results and the physiological data of the subjects in the training samples are input into the multimodal data detection OSA model to obtain the OSA disease severity of the subjects corresponding to the training samples, and the loss value of the multimodal data detection OSA model is calculated according to the OSA disease severity of the subjects corresponding to each training sample and the OSA category label corresponding to the subjects. The loss value is used to update the parameters of the multimodal data detection OSA model, and the successfully trained multimodal data detection OSA model is obtained after the training is completed.

[0144] In an embodiment of the present invention, the loss value is calculated using the OSA disease severity output by the multimodal data detection OSA model and the OSA category label of the subject, and then the multimodal data detection OSA model is updated in reverse. By continuously training according to this step, a successfully trained multimodal data detection OSA model can be obtained, which provides basic model support for obtaining a successfully trained obstructive sleep apnea detection model.

[0145] The obstructive sleep apnea detection device provided by the present invention is described below. The obstructive sleep apnea detection device described below and the obstructive sleep apnea detection method described above can be referred to each other.

[0146] Figure 5 A schematic diagram of the structure of an obstructive sleep apnea detection device provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, an embodiment of the present invention provides an obstructive sleep apnea detection device, comprising:

[0147] The data acquisition module 501 is used to acquire the physiological data of the person being tested and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal and a pulse wave signal;

[0148] The signal processing module 502 is used to perform slicing and filtering on the sleep monitoring signal to obtain at least one frame of slice signal;

[0149] The result acquisition module 503 is used to input the physiological data and the at least one frame of slice signal into the successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the tested person;

[0150] The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model and a multimodal data detection OSA model.

[0151] The obstructive sleep apnea detection device provided by the present invention first obtains the physiological data of the person to be tested and a group of sleep monitoring signals in a preset time period, then slices and filters the sleep monitoring signals to obtain multiple frames of sliced ​​signals, and finally inputs the multiple frames of sliced ​​signals into a successfully trained obstructive sleep apnea detection model to automatically predict the OSA severity of the person to be tested, thereby reducing a large amount of manual intervention in the OSA detection process, improving detection efficiency, and at the same time improving accuracy. Moreover, since each detection uses the same successfully trained obstructive sleep apnea detection model, the stability of the detection result is also improved to a certain extent.

[0152] Optionally, the result acquisition module 503 is used to:

[0153] Inputting the at least one frame of slice signal into a successfully trained sleep stage classification model to obtain a sleep stage high-level feature vector corresponding to each frame of the slice signal;

[0154] Inputting the at least one frame of slice signal into a successfully trained respiratory event classification model to obtain a respiratory event classification high-level feature vector corresponding to each frame of the slice signal;

[0155] Inputting the sleep stage high-level feature vector corresponding to each frame of the slice signal and the respiratory event classification high-level feature vector corresponding to each frame of the slice signal into the successfully trained high-level sample classification model to obtain the feature vector corresponding to each frame of the slice signal;

[0156] Integrate the characteristic vectors of all the slice signals, and calculate the proportion of all slice signals that are in a sleeping state and have respiratory events in all slice signals;

[0157] The proportion result and the physiological data are input into a successfully trained multimodal data OSA detection model to obtain the severity of obstructive sleep apnea suffered by the tested person.

[0158] Optionally, the device further comprises a model training module, and the model training module is used to train the obstructive sleep apnea detection model;

[0159] The training of the obstructive sleep apnea detection model comprises:

[0160] Build an obstructive sleep apnea detection model;

[0161] Acquire a set of sleep monitoring signal samples of at least one subject, a sleep stage label corresponding to the subject, a respiratory event label corresponding to the subject, an OSA category label corresponding to the subject, and physiological data of the subject;

[0162] Performing slicing and filtering processing on the sleep monitoring signal samples of each subject to obtain at least one frame of slice signal samples;

[0163] According to the sleep stage label corresponding to each subject and the respiratory event label corresponding to the subject, obtaining the sleep stage label and the corresponding respiratory event label corresponding to each frame of the slice signal sample;

[0164] Constructing a training sample set; wherein the training sample set includes at least one training sample, each of the training samples includes all slice signal samples of a subject, physiological data of the subject, a sleep stage label corresponding to each frame of the slice signal sample, a respiratory event label corresponding to each frame of the slice signal sample, and an OSA category label corresponding to the subject;

[0165] The obstructive sleep apnea detection model is trained using the training sample set.

[0166] Optionally, the using the training sample set to train the obstructive sleep apnea detection model includes:

[0167] Training the sleep stage classification model according to all the slice signal samples in the training sample and the sleep stage labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained sleep stage classification model after the training is completed;

[0168] Training the respiratory event classification model according to all the slice signal samples in the training sample and the respiratory event labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained respiratory event classification model after the training is completed;

[0169] Input all the slice signal samples in the training samples into the successfully trained sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal samples; input all the slice signal samples in the training samples into the successfully trained respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples;

[0170] According to the sleep staging high-level feature samples and respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples, the sleep staging labels corresponding to each frame of the slice signal samples, and the respiratory event labels corresponding to each frame of the slice signal samples, the high-level sample classification model is trained to obtain a successfully trained high-level sample classification model after the training is completed;

[0171] Input the sleep stage high-level feature samples corresponding to each frame of the slice signal samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples into the successfully trained high-level sample classification model to obtain feature vectors of all the slice signal samples;

[0172] Integrate the feature vectors of all the slice signal samples, and calculate the proportion of all slice signal samples that are in a sleeping state and have respiratory events in all slice signal samples;

[0173] The multimodal data OSA detection model is trained according to the proportion results, the physiological data of the subject in the training sample, and the OSA category label corresponding to the subject, and the successfully trained multimodal data OSA detection model is obtained after the training is completed.

[0174] Optionally, the training of the sleep stage classification model includes:

[0175] Inputting all the slice signal samples in the training sample into the sleep stage classification model to obtain sleep stage high-level feature samples corresponding to each frame of the slice signal sample;

[0176] According to the sleep stage high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the sleep stage labels corresponding to the slice signal samples in each frame, the loss value of the sleep stage classification model is calculated, and the parameters of the sleep stage classification model are updated using the loss value.

[0177] Optionally, inputting all the slice signal samples in the training samples into the sleep stage classification model to obtain sleep stage high-level feature samples corresponding to each frame of the slice signal samples includes:

[0178] Inputting all the slice signal samples into the neural network of the sleep stage classification model, and extracting high-level representation samples of each frame of the slice signal samples;

[0179] Inputting the high-level representation samples of the slice signal samples of each frame into multiple inverted residual modules with grouped attention mechanisms of the sleep stage classification model, and performing density clustering on the high-level representation samples belonging to the same type of OSA samples to obtain a high-level representation subset of each type of OSA;

[0180] The high-level representation subset of each type of OSA is input into the average pooling layer and the fully connected layer of the sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to the slice signal samples of each frame.

[0181] Optionally, the training of the respiratory event classification model comprises:

[0182] Inputting all the slice signal samples in the training sample into the respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples;

[0183] According to the respiratory event classification high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the respiratory event labels corresponding to the slice signal samples in each frame, the loss value of the respiratory event classification model is calculated, and the parameters of the respiratory event classification model are updated using the loss value.

[0184] Optionally, the training of the high-level sample classification model includes:

[0185] Input the sleep stage high-level feature samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples into the high-level sample classification model to obtain a feature vector of each frame of the slice signal samples;

[0186] According to the feature vector of the slice signal sample in each frame of each training sample, the sleep stage label corresponding to the slice signal sample in each frame, and the respiratory event label corresponding to the slice signal sample in each frame, the loss value of the high-level sample classification model is calculated, and the parameters of the high-level sample classification model are updated using the loss value.

[0187] Optionally, the training of the multimodal data OSA detection model includes:

[0188] The proportion results and the physiological data of the subjects in the training samples are input into the multimodal data detection OSA model to obtain the OSA disease severity of the subjects corresponding to the training samples, and the loss value of the multimodal data detection OSA model is calculated according to the OSA disease severity of the subjects corresponding to each training sample and the OSA category label corresponding to the subjects. The loss value is used to update the parameters of the multimodal data detection OSA model, and the successfully trained multimodal data detection OSA model is obtained after the training is completed.

[0189] It should be noted here that the above-mentioned obstructive sleep apnea detection device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned obstructive sleep apnea detection device embodiment, and can achieve the same technical effects. The parts and beneficial effects that are the same as those in the method embodiment will not be specifically repeated here.

[0190] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the obstructive sleep apnea detection method, which includes: obtaining physiological data of the person under test and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal and a pulse wave signal; performing slicing and filtering on the sleep monitoring signal to obtain at least one frame of slice signal; inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome OSA of the person under test; wherein the obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model and a multimodal data detection OSA model.

[0191] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0192] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the obstructive sleep apnea detection method provided by the above-mentioned methods, the method comprising: obtaining physiological data of a person under test and a set of sleep monitoring signals in a preset time period, the sleep monitoring signal comprising at least one of an electrocardiogram signal, a blood oxygen signal and a pulse wave signal; performing slicing and filtering processing on the sleep monitoring signal to obtain at least one frame of slice signal; inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the person under test; wherein the obstructive sleep apnea detection model comprises a sleep stage classification model, a respiratory event classification model, a high-level sample classification model and a multimodal data detection OSA model.

[0193] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0194] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting obstructive sleep apnea, characterized in that: include: Acquire physiological data of the person being measured and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal, and a pulse wave signal; Performing slicing and filtering processing on the sleep monitoring signal to obtain at least one frame of slice signal; Inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the tested person; The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multimodal data detection OSA model; The step of inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the person being tested includes: Inputting the at least one frame of slice signal into a successfully trained sleep stage classification model to obtain a sleep stage high-level feature vector corresponding to each frame of the slice signal; Inputting the at least one frame of slice signal into a successfully trained respiratory event classification model to obtain a respiratory event classification high-level feature vector corresponding to each frame of the slice signal; Inputting the sleep stage high-level feature vector corresponding to each frame of the slice signal and the respiratory event classification high-level feature vector corresponding to each frame of the slice signal into the successfully trained high-level sample classification model to obtain the feature vector corresponding to each frame of the slice signal; Integrate the feature vectors of all slice signals processed by the successfully trained high-level sample classification model, and calculate the proportion of all slice signals in the sleep state and with respiratory events in all slice signals; The proportion result and the physiological data are input into a successfully trained multimodal data OSA detection model to obtain the severity of obstructive sleep apnea suffered by the tested person.

2. The obstructive sleep apnea detection method according to claim 1, characterized in that: The method further comprises: training the obstructive sleep apnea detection model; The training of the obstructive sleep apnea detection model comprises: Build an obstructive sleep apnea detection model; Acquire a set of sleep monitoring signal samples of at least one subject, a sleep stage label corresponding to the subject, a respiratory event label corresponding to the subject, an OSA category label corresponding to the subject, and physiological data of the subject; Performing slicing and filtering processing on the sleep monitoring signal samples of each subject to obtain at least one frame of slice signal samples; According to the sleep stage label corresponding to each subject and the respiratory event label corresponding to the subject, obtaining the sleep stage label and the corresponding respiratory event label corresponding to each frame of the slice signal sample; Constructing a training sample set; wherein the training sample set includes at least one training sample, each of the training samples includes all slice signal samples of a subject, physiological data of the subject, a sleep stage label corresponding to each frame of the slice signal sample, a respiratory event label corresponding to each frame of the slice signal sample, and an OSA category label corresponding to the subject; The obstructive sleep apnea detection model is trained using the training sample set.

3. The obstructive sleep apnea detection method according to claim 2, characterized in that: The step of training the obstructive sleep apnea detection model using the training sample set includes: Training the sleep stage classification model according to all the slice signal samples in the training sample and the sleep stage labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained sleep stage classification model after the training is completed; Training the respiratory event classification model according to all the slice signal samples in the training sample and the respiratory event labels corresponding to each frame of the slice signal samples, and obtaining a successfully trained respiratory event classification model after the training is completed; Input all the slice signal samples in the training samples into the successfully trained sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal samples; input all the slice signal samples in the training samples into the successfully trained respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples; According to the sleep staging high-level feature samples and respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples, the sleep staging labels corresponding to each frame of the slice signal samples, and the respiratory event labels corresponding to each frame of the slice signal samples, the high-level sample classification model is trained to obtain a successfully trained high-level sample classification model after the training is completed; Input the sleep stage high-level feature samples corresponding to each frame of the slice signal samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples into the successfully trained high-level sample classification model to obtain feature vectors of all the slice signal samples; Integrate the feature vectors of all the slice signal samples, and calculate the proportion of all slice signal samples that are in a sleeping state and have respiratory events in all slice signal samples; The multimodal data OSA detection model is trained according to the proportion results, the physiological data of the subject in the training sample, and the OSA category label corresponding to the subject, and the successfully trained multimodal data OSA detection model is obtained after the training is completed.

4. The obstructive sleep apnea detection method according to claim 3, wherein the training of the sleep stage classification model comprises: Inputting all the slice signal samples in the training sample into the sleep stage classification model to obtain sleep stage high-level feature samples corresponding to each frame of the slice signal sample; According to the sleep stage high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the sleep stage labels corresponding to the slice signal samples in each frame, the loss value of the sleep stage classification model is calculated, and the parameters of the sleep stage classification model are updated using the loss value.

5. The obstructive sleep apnea detection method according to claim 4, characterized in that: The step of inputting all the slice signal samples in the training sample into the sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to each frame of the slice signal sample comprises: Inputting all the slice signal samples into the neural network of the sleep stage classification model, and extracting high-level representation samples of each frame of the slice signal samples; Inputting the high-level representation samples of the slice signal samples of each frame into multiple inverted residual modules with grouped attention mechanisms of the sleep stage classification model, and performing density clustering on the high-level representation samples belonging to the same type of OSA samples to obtain a high-level representation subset of each type of OSA; The high-level representation subset of each type of OSA is input into the average pooling layer and the fully connected layer of the sleep stage classification model to obtain the sleep stage high-level feature samples corresponding to the slice signal samples of each frame.

6. The obstructive sleep apnea detection method according to claim 3, wherein the training of the respiratory event classification model comprises: Inputting all the slice signal samples in the training sample into the respiratory event classification model to obtain the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples; According to the respiratory event classification high-level feature samples corresponding to the slice signal samples in each frame of each training sample and the respiratory event labels corresponding to the slice signal samples in each frame, the loss value of the respiratory event classification model is calculated, and the parameters of the respiratory event classification model are updated using the loss value.

7. The obstructive sleep apnea detection method according to claim 3, wherein the training of the high-level sample classification model comprises: Input the sleep stage high-level feature samples and the respiratory event classification high-level feature samples corresponding to each frame of the slice signal samples into the high-level sample classification model to obtain a feature vector of each frame of the slice signal samples; According to the feature vector of the slice signal sample in each frame of each training sample, the sleep stage label corresponding to the slice signal sample in each frame, and the respiratory event label corresponding to the slice signal sample in each frame, the loss value of the high-level sample classification model is calculated, and the parameters of the high-level sample classification model are updated using the loss value.

8. The obstructive sleep apnea detection method according to claim 3, wherein the training of the multimodal data OSA detection model comprises: The proportion results and the physiological data of the subjects in the training samples are input into the multimodal data detection OSA model to obtain the OSA disease severity of the subjects corresponding to the training samples, and the loss value of the multimodal data detection OSA model is calculated according to the OSA disease severity of the subjects corresponding to each training sample and the OSA category label corresponding to the subjects. The loss value is used to update the parameters of the multimodal data detection OSA model, and the successfully trained multimodal data detection OSA model is obtained after the training is completed.

9. An obstructive sleep apnea detection device, characterized in that: include: A data acquisition module, used to acquire physiological data of the person being tested and a set of sleep monitoring signals in a preset time period, wherein the sleep monitoring signal includes at least one of an electrocardiogram signal, a blood oxygen signal, and a pulse wave signal; A signal processing module, used for performing slicing and filtering on the sleep monitoring signal to obtain at least one frame of slice signal; A result acquisition module is used to input the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the tested person; The obstructive sleep apnea detection model includes a sleep stage classification model, a respiratory event classification model, a high-level sample classification model, and a multimodal data detection OSA model; The step of inputting the physiological data and the at least one frame of slice signal into a successfully trained obstructive sleep apnea detection model to obtain the degree of obstructive sleep apnea hypopnea syndrome (OSA) of the person being tested includes: Inputting the at least one frame of slice signal into a successfully trained sleep stage classification model to obtain a sleep stage high-level feature vector corresponding to each frame of the slice signal; Inputting the at least one frame of slice signal into a successfully trained respiratory event classification model to obtain a respiratory event classification high-level feature vector corresponding to each frame of the slice signal; Inputting the sleep stage high-level feature vector corresponding to each frame of the slice signal and the respiratory event classification high-level feature vector corresponding to each frame of the slice signal into the successfully trained high-level sample classification model to obtain the feature vector corresponding to each frame of the slice signal; Integrate the feature vectors of all slice signals processed by the successfully trained high-level sample classification model, and calculate the proportion of all slice signals in the sleep state and with respiratory events in all slice signals; The proportion result and the physiological data are input into a successfully trained multimodal data OSA detection model to obtain the severity of obstructive sleep apnea suffered by the tested person.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the obstructive sleep apnea detection method according to any one of claims 1 to 8 is implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the obstructive sleep apnea detection method according to any one of claims 1 to 8 is implemented.

Images