Physiological data processing method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a physiological data processing method and device, equipment and a storage medium. The method provided by the embodiment of the invention comprises the following steps: receiving a physiological data set for sleep monitoring from detection equipment; identifying a machine learning model corresponding to the physiological data set, the physiological data set having a first channel associated with a data acquisition path of the detection device, the corresponding machine learning model being configured to process physiological data having a standard channel; determining a mapping relationship between the first channel and the standard channel, and mapping the first channel to the standard channel based on the mapping relationship; and inputting the mapped physiological data in the physiological data set into the corresponding machine learning model for processing.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and more specifically, to a method, apparatus, device, and storage medium for processing physiological data. Background Technology

[0002] Obstructive sleep apnea-hypopnea syndrome (OSAHS or OSA) is a chronic disease whose health hazards are severely underestimated. It is characterized by recurrent partial or complete collapse of the upper airway during sleep, causing frequent apnea and hypopnea events, leading to intermittent hypoxia and fragmented sleep, resulting in damage to a range of target organs, including hypertension, coronary artery disease, arrhythmias, pulmonary heart disease, and heart failure. According to the Global Burden of Disease study published in *Lancet Respiratory Medicine* in 2019, the number of people with obstructive sleep apnea-hypopnea syndrome (OSAHS) in my country is as high as 176 million, of whom more than 66 million require active treatment. Furthermore, with changes in lifestyle and population aging, the incidence of OSAHS is also on the rise, and its health hazards have become a significant public health issue that urgently needs to be addressed. OSAHS users often experience neuroendocrine dysfunction due to repeated hypoxia, leading to daytime sleepiness, memory loss, mood instability, and reduced work efficiency, which seriously affects their quality of life.

[0003] However, the development of my country's sleep medicine service system started relatively late. Although significant progress has been made in the construction of sleep centers in tertiary hospitals in recent years, insufficient resource allocation still exists in primary healthcare institutions. Data released in 2023 by the Sleep Medicine Branch of the China International Exchange and Promotion Association for Medical and Health Care, covering nearly 600 primary, secondary, and tertiary hospitals, shows that less than 50% of secondary and tertiary hospitals have sleep centers equipped with the gold standard polysomnography (PSG), while only 2% of primary hospitals have such centers. This translates to less than 0.2 sleep centers per 100,000 people, indicating a severe shortage of clinical medical resources. Furthermore, according to the survey results, the primary reason why primary hospitals have failed to provide sleep medicine services as needed is a severe shortage of professional sleep technicians. Globally, there are approximately 20,000 certified Registered Polysomnography Technicians (RPSGT), while my country has fewer than 300, far below clinical demand. The other two top reasons are the high cost of the gold standard PSG equipment and the lack of professional training services. This resource imbalance directly restricts the diagnostic and treatment capabilities of primary healthcare institutions, increasing the burden on patients seeking medical care across regions and exacerbating the imbalance in the allocation of medical resources.

[0004] Furthermore, sleep monitoring devices from different manufacturers may have different types of sensors, varying numbers and types of channels, and different data formats, resulting in completely different physiological data output from different devices. Existing models for processing physiological data have specific requirements for the input channels; for example, some models require physiological data with standard channels. This makes existing models incompatible with sleep monitoring devices from different manufacturers, limiting their usability and hindering accurate results. In addition, signal noise, abnormal data, and missing channels can also degrade the model's detection results.

[0005] Therefore, it is desirable to have a method, apparatus, device, and storage medium that can perform channel mapping and preprocessing of physiological data output from sleep monitoring devices produced by different manufacturers, which can be effectively compatible with different device sensor types, channel types, and data formats, and improve the robustness of the model by masking and reconstructing the training data during the training of the model used to process the physiological data, thereby maximizing its clinical applicability. Summary of the Invention

[0006] Embodiments of this disclosure provide a method, apparatus, device, and storage medium for processing physiological data for sleep monitoring.

[0007] Embodiments of this disclosure provide a method for processing physiological data for sleep monitoring, comprising: receiving a physiological dataset for sleep monitoring from a detection device; identifying a machine learning model corresponding to the physiological dataset, wherein the physiological dataset has a first channel, the first channel being associated with a data acquisition path of the detection device, and the corresponding machine learning model being configured to process physiological data having a standard channel; determining a mapping relationship between the first channel and the standard channel, mapping the first channel to the standard channel based on the mapping relationship; and inputting physiological data from the mapped physiological dataset into the corresponding machine learning model for processing.

[0008] According to an embodiment of the present disclosure, mapping the first channel to the standard channel based on the mapping relationship includes: determining whether the first channel has a missing channel compared to the standard channel, and filling the missing channel in response to the first channel having a missing channel compared to the standard channel.

[0009] According to an embodiment of the present disclosure, the method of filling the missing channel includes: determining whether there are other channels besides the standard channel in the first channel compared to the standard channel, and filling the missing channel with the other channels in response to the existence of the other channels.

[0010] According to an embodiment of the present disclosure, the method of filling the missing channel includes: determining whether there are other channels besides the standard channel in the first channel compared to the standard channel, and filling the missing channel with a mask in response to the absence of the other channels.

[0011] According to an embodiment of the present disclosure, mapping the first channel to the standard channel based on the mapping relationship includes: determining whether the first channel has a missing channel compared to the standard channel and determining whether the first channel has other channels besides the standard channel compared to the standard channel; and discarding the other channels in response to the first channel not having a missing channel compared to the standard channel and having the other channels.

[0012] According to the method of the embodiments of this disclosure, the process of inputting physiological data from the mapped physiological dataset into the corresponding machine learning model for processing further includes: selecting a machine learning model corresponding to the physiological dataset from one or more machine learning models, inputting physiological data from the mapped physiological dataset into the corresponding machine learning model, and processing the physiological data from the mapped physiological dataset using the corresponding machine learning model.

[0013] According to the method of an embodiment of the present disclosure, the type of the detection device includes a contact sleep detection device, and the selection of a machine learning model corresponding to the physiological dataset includes selecting a first machine learning model corresponding to the contact sleep detection device from one or more machine learning models, wherein the first machine learning model processes at least electrocardiogram data, chest strap data, and abdominal strap data.

[0014] According to the method of embodiments of the present disclosure, the first machine learning model further processes at least one or more of the following: electroencephalogram (EEG) data, eye movement data, electromyography (EMG) data, airflow channel data, nasal pressure channel data, transcutaneous oxygen saturation channel data, and snoring channel data.

[0015] According to the method of embodiments of the present disclosure, the contact sleep detection device includes at least one of a polysomnography monitor and a home sleep apnea monitor.

[0016] According to an embodiment of the method of this disclosure, wherein the type of detection device includes a non-contact sleep detection device, and wherein selecting a machine learning model corresponding to the physiological dataset includes selecting a second machine learning model corresponding to the non-contact sleep detection device from one or more machine learning models, the second machine learning model processing at least heart rate data and breathing data.

[0017] According to the method of an embodiment of the present disclosure, the second machine learning model further processes at least one of snoring data and body movement data.

[0018] According to the method of embodiments of this disclosure, the one or more machine learning models are trained based on physiological data with standard channels.

[0019] According to the method of embodiments of this disclosure, the physiological data having a standard channel is subjected to one or more of random white noise, amplitude stretching, and frequency offset.

[0020] According to the method of an embodiment of the present disclosure, the physiological data having a standard channel is subjected to one or more of the following operations: masking and reconstructing physiological data at a single time point of a single channel of the physiological data having a standard channel with a predetermined probability; masking and reconstructing physiological data at the same time point of all channels of the physiological data having a standard channel; masking and reconstructing physiological data for a complete time of a predetermined channel of the physiological data having a standard channel; and masking and reconstructing physiological data for a predetermined time period of adjacent channels of the physiological data having a standard channel.

[0021] According to the method of the embodiments of this disclosure, mapping the first channel to the standard channel based on the mapping relationship further includes: converting the name of the physiological data in the physiological dataset, converting the encoding method of the physiological data in the physiological dataset, converting the sampling rate of the physiological data in the physiological dataset, converting the unit of the physiological data in the physiological dataset, and converting the data type of the physiological data in the physiological dataset.

[0022] The method according to an embodiment of this disclosure further includes, in response to the presence of a missing channel in the first channel of the physiological data in the physiological dataset compared to the standard channel, increasing the weight in the corresponding machine learning model corresponding to the mapped first channel.

[0023] The method according to embodiments of this disclosure further includes: removing abnormal data from the physiological dataset by means of signal energy threshold detection and morphological analysis.

[0024] According to the method of embodiments of the present disclosure, the one or more machine learning models include a first machine learning model corresponding to a contact sleep detection device in a sleep detection device and a second machine learning model corresponding to a non-contact sleep detection device in a sleep detection device, wherein the second machine learning model is obtained from the first machine learning model based on transfer learning.

[0025] According to the method of the embodiments of this disclosure, the process of inputting physiological data from the mapped physiological dataset into the corresponding machine learning model for processing includes: the corresponding machine learning model identifying at least one of sleep stages, respiratory events, and apnea-hypopnea index based on the mapped physiological dataset.

[0026] According to the method of embodiments of the present disclosure, the sleep stages include at least one of wakefulness, light sleep, deep sleep, and REM sleep, and the breathing events include at least obstructive breathing events.

[0027] According to an embodiment of the present disclosure, the method further includes preprocessing the physiological data in the physiological dataset, wherein the preprocessing includes one or more of the following: performing time-frequency plot calculation on the physiological data in the physiological dataset; extracting R-peaks from the physiological data in the physiological dataset; and performing bandpass filtering on the physiological data in the physiological dataset.

[0028] Embodiments of this disclosure provide an apparatus for processing physiological data for sleep monitoring. The apparatus includes: a receiving module configured to receive a physiological dataset for sleep monitoring from a detection device; an identification module configured to identify a machine learning model corresponding to the physiological dataset, wherein the physiological dataset has a first channel, the first channel being associated with a data acquisition path of the detection device, and the corresponding machine learning model being configured to process physiological data having a standard channel; a mapping module configured to determine a mapping relationship between the first channel and the standard channel, and map the first channel to the standard channel based on the mapping relationship; and a processing module for inputting the physiological data in the mapped physiological dataset into the corresponding machine learning model for processing.

[0029] Embodiments of this disclosure provide an apparatus for processing physiological data, comprising: one or more processors; and one or more memories storing a computer-executable program that, when executed by the processor, performs the method described above.

[0030] Embodiments of this disclosure provide a computer program product, including a computer program or instructions, wherein the computer program or instructions implement the above-described method when executed by a processor.

[0031] Embodiments of this disclosure provide a computer-readable storage medium having computer-executable instructions stored thereon, which, when executed by a processor, are used to implement the methods described above.

[0032] The methods, apparatus, devices, and storage media for processing physiological data provided in the embodiments of this disclosure are compatible with various models of contact and non-contact sleep monitoring devices from different manufacturers. In this way, the channel differences in physiological data caused by the different models and types of sleep monitoring devices from different manufacturers can be eliminated, thereby avoiding limitations on the performance of models for processing physiological data.

[0033] The method provided according to the embodiments of this disclosure can improve its versatility for clinical application. Because the method according to the embodiments of this disclosure is compatible with different models and types of sleep monitoring devices from different manufacturers, it avoids the need for medical institutions to repeatedly install different models for processing physiological data on sleep monitoring devices from different manufacturers, and to repeatedly perform the corresponding security certification process. Furthermore, due to its strong compatibility, the method according to the embodiments of this disclosure avoids the need for extensive training of medical personnel on different models for processing physiological data. Therefore, it not only improves the diagnostic capabilities of primary healthcare institutions for sleep disorders, but also helps to achieve the downward flow of medical resources and improve the balance of regional medical services.

[0034] Furthermore, the method provided according to the embodiments of this disclosure can alleviate the problem of insufficient allocation of primary healthcare resources. Since the method according to the embodiments of this disclosure is compatible with different models of home sleep apnea monitors and non-contact sleep testing devices from different manufacturers, users can perform sleep disorder testing at home, reducing the burden of cross-regional medical treatment and alleviating the problem of insufficient allocation of medical resources. At the same time, due to its strong compatibility, the method according to the embodiments of this disclosure can reduce the operational complexity of home sleep disorder testing and improve user compliance with home-based sleep disorder testing. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some exemplary embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0036] Figure 1 This is a flowchart illustrating a method for processing physiological data according to an embodiment of the present disclosure;

[0037] Figure 2 This is a flowchart illustrating a method for processing physiological data according to an embodiment of the present disclosure;

[0038] Figure 3A A schematic diagram of a non-contact sleep detection device according to an embodiment of the present disclosure is shown;

[0039] Figure 3B This is an example scenario of processing physiological data according to an embodiment of the present disclosure;

[0040] Figure 3C This is an example scenario of processing physiological data according to an embodiment of the present disclosure;

[0041] Figure 3D This is an example scenario of preprocessing physiological data according to an embodiment of the present disclosure;

[0042] Figure 3E This is an example scenario of processing physiological data with a standard channel according to an embodiment of the present disclosure;

[0043] Figure 3F This is an example scenario of masking and reconstructing physiological data used to train a machine learning model according to an embodiment of the present disclosure;

[0044] Figure 4 This is a schematic diagram of an apparatus for processing physiological data according to an embodiment of the present disclosure;

[0045] Figure 5 A schematic diagram of a device 500 for processing physiological data according to an embodiment of the present disclosure is shown; and

[0046] Figure 6 A schematic diagram of the architecture of an exemplary computing device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0048] In this specification and accompanying drawings, steps and elements that are substantially the same or similar are indicated by the same or similar reference numerals, and repeated descriptions of these steps and elements are omitted. Furthermore, in the description of this disclosure, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance or order.

[0049] In embodiments of this disclosure, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0051] To facilitate the description of this disclosure, the following concepts related to this disclosure are introduced.

[0052] The method for processing physiological data disclosed herein may involve polysomnography (PSG). PSG is a medical examination method used to comprehensively assess a user's sleep quality and identify sleep disorders. PSG learns about a user's sleep structure, cycles, the characteristics of each stage, and the duration of sleep by simultaneously recording and analyzing physiological data from multiple sleep periods. PSG is an objective examination for analyzing sleep structure and assessing sleep disorders; it is a fundamental evaluation technique in sleep medicine and the "gold standard" for diagnosing sleep disorders. In medical diagnosis, the "gold standard" refers to a widely recognized, most accurate, and reliable diagnostic method or testing tool. However, due to the comprehensiveness of the diagnosis, PSG has high requirements for the monitoring environment, complex wiring, numerous physiological data acquisition leads, and requires qualified technicians to operate. Subjects must spend the night in the sleep laboratory in the presence of technicians, which can lead to the "first-night effect," resulting in measurement results that lack authenticity. The "first-night effect" refers to the fact that most users experience worse sleep quality on the first night of sleep monitoring in the sleep monitoring room compared to subsequent nights. Generally, compared to continuous sleep monitoring over more than one night, the false negative rate for OSA (Obstructive Sleep Apnea) during a single night's sleep monitoring can reach 55%, and relying solely on a single night's diagnosis may underestimate the severity of OSA. Furthermore, the contact-based fixation and connection method between the physiological data acquisition components of a sleep physiotherapy (PSG) device and the user is complex, making it highly susceptible to detachment or poor contact during sleep due to factors such as movement and sweating, leading to incomplete data or interference. PSG devices are expensive, and equipment resources are scarce, potentially resulting in long waiting times for appointments and high examination costs.

[0053] The method for processing physiological data disclosed herein can relate to a home sleep apnea monitor for performing home sleep testing (HSAT). The home sleep apnea monitor can be a portable device for monitoring and diagnosing obstructive sleep apnea (OSA) at home. Home sleep apnea monitors collect less physiological data than PSGs and have simpler wiring. However, the contact-based fixation and connection method of the physiological data acquisition components of a home sleep apnea monitor makes the components highly susceptible to detachment or poor contact during the user's sleep due to factors such as movement and sweating, leading to incomplete or interfering data. While home sleep apnea monitors offer advantages over PSGs in terms of lower cost and convenience, the uncertainty of their results and the potential need for retesting can prolong the user's medical visits, increase the disease burden, and reduce user adherence to treatment.

[0054] The embodiments of this disclosure will now be described in further detail with reference to the accompanying drawings.

[0055] Figure 1 This is a flowchart illustrating a method for processing physiological data for sleep monitoring according to embodiments of the present disclosure. Figure 1 As shown, method 100 may include steps S110-S150.

[0056] In step S110, a physiological dataset for sleep monitoring can be received from the detection device. According to one embodiment of this disclosure, the detection device may include a contact sleep monitoring device and a non-contact sleep monitoring device. For example, a physiological dataset including one or more physiological data can be received from a contact sleep monitoring device such as a PSG or a home sleep apnea monitor. The physiological dataset may include one or more of the following: electroencephalogram (EEG) data, eye movement data, electromyography (EMG) data, electrocardiogram (ECG) data, nasopharyngeal airflow data, respiratory effort and pattern data, and blood oxygen saturation data, heart rate data, respiratory data, snoring data, and body movement data.

[0057] In step S120, a machine learning model corresponding to the physiological dataset can be identified. In one embodiment, the physiological dataset has a first channel, which is associated with the data acquisition path of the detection device, and the corresponding machine learning model is configured to process physiological data with a standard channel. For example, the machine learning model may only be able to process physiological data with a standard channel. There may be a correspondence between the machine learning model, the physiological dataset, and the type of detection device. For example, there may be two machine learning models as examples, where the standard channel corresponding to the first machine learning model may be the first standard channel, and the standard channel corresponding to the second machine learning model may be the second standard channel. According to an embodiment of this disclosure, the physiological data that the first machine learning model can process may include at least electrocardiogram data and respiratory effort and pattern data. Additionally or alternatively, the physiological data that the first machine learning model can process may also include one or more of electroencephalogram data, eye movement data, electromyogram data, nasopharyngeal airflow data, and blood oxygen saturation data. The physiological data that the second machine learning model can process may include at least heart rate data and respiratory data. Additionally or alternatively, the physiological data that the second machine learning model can process may also include at least one of snoring data and body movement data. The physiological data that the first machine learning model can process and the physiological data that the second machine learning model can process are exemplary and not limiting. The physiological data that the first machine learning model can process and the physiological data that the second machine learning model can process may include more or less content, and this disclosure is not limited thereto.

[0058] According to embodiments of this disclosure, the type of detection device corresponding to the physiological dataset and the machine learning model to be used can be identified by the physical characteristics of the physiological dataset. The first channel can be related to the data acquisition path of the detection device. For example, the first channel can be related to the model and manufacturer of the detection device; that is, the first channel of physiological data detected by detection devices of different models and manufacturers can have significant differences. The corresponding machine learning model can be related to a standard channel; for example, the corresponding machine learning model can only, or can only, process physiological data with a standard channel with superior performance.

[0059] In step S130, a mapping relationship between the first channel and the standard channel can be determined, and the first channel is mapped to the standard channel based on the mapping relationship. A channel can refer to an independent data acquisition path or signal input port. The first channel and the standard channel can each include one or more channels. According to one embodiment of this disclosure, the first channel is associated with the data acquisition path of the detection device. For example, the detection device receiving physiological datasets from it can be a detection device with a first channel. According to one embodiment of this disclosure, the standard channel can be associated with a corresponding machine learning model. For example, the corresponding machine learning model can process physiological data corresponding to the standard channel. The first channel of the physiological data in the physiological dataset may have a mapping relationship with the standard channel of the physiological data that the corresponding machine learning model can process, but the first channel and the standard channel can be different. For example, compared to the standard channel, the first channel may have missing channels or redundant channels. Missing channels may be due to the physiological data acquisition components (e.g., electrodes) of contact sleep monitoring devices such as PSGs or home sleep apnea monitors detaching from the user (e.g., the user's skin), or it may be due to differences in the model of the monitoring device (e.g., different models of PSGs, home sleep apnea monitors, and non-contact sleep monitoring devices may correspond to different channels), thus resulting in the loss of physiological data for some channels. Excessive channels may be due to configuration issues by the monitoring device manufacturer.

[0060] According to one embodiment of this disclosure, mapping a first channel to a standard channel based on a mapping relationship includes determining whether the first channel has a missing channel compared to the standard channel. In response to the presence of a missing channel in the first channel compared to the standard channel, the missing channel can be filled. By filling the missing channel, the first channel can be matched with the standard channel.

[0061] In one embodiment, for example, the first channel may include channels numbered [1,2,4,5,6,7,8], and the standard channel may include channels numbered [1,2,3,4,5,6,7]. Therefore, the first channel may have a missing channel compared to the standard channel, i.e., channel number 3. Furthermore, the first channel may have other channels besides the standard channel, i.e., channel number 8. In this case, filling the missing channel may include determining whether the first channel has other channels besides the standard channel. In response to the existence of other channels (i.e., channel number 8), the missing channel (i.e., channel number 3) is filled using the other channels (i.e., channel number 8).

[0062] In one embodiment, for example, the first channel may include channels numbered [1,2,4,5,6,7], and the standard channel may include channels numbered [1,2,3,4,5,6,7]. Therefore, the first channel may have a missing channel compared to the standard channel, i.e., channel number 3. Furthermore, the first channel may not have any other channels besides the standard channel. In this case, filling the missing channel may include determining whether the first channel has any other channels besides the standard channel. In response to the absence of other channels, a mask can be used to fill the missing channel (i.e., channel number 3).

[0063] In one embodiment, for example, the first channel may include channels numbered [1,2,3,4,5,6,7,8], and the standard channel may include channels numbered [1,2,3,4,5,6,7]. Therefore, the first channel may not have any missing channels compared to the standard channel. Furthermore, the first channel may have other channels besides the standard channel, i.e., channel number 8. In this case, mapping the first channel to the standard channel based on the mapping relationship may include determining whether the first channel has any missing channels compared to the standard channel and determining whether the first channel has other channels besides the standard channel. In response to the first channel not having any missing channels compared to the standard channel and the existence of other channels (i.e., channel number 8), the other channel (i.e., channel number 8) may be discarded.

[0064] In step S140, the physiological data in the mapped physiological dataset can be input into the corresponding machine learning model for processing. According to an embodiment of this disclosure, even if the physiological dataset has missing or redundant channels compared to the standard channels associated with the corresponding machine learning model, by mapping the first channel of the physiological data in the physiological dataset, the corresponding machine learning model can still process the physiological dataset with better performance. By mapping the physiological data channels, the method according to the embodiments of this disclosure can be compatible with various models of contact and non-contact sleep monitoring devices produced by different manufacturers. In this way, the performance limitations of the model for processing physiological data caused by the different models and types of sleep monitoring devices produced by different manufacturers can be eliminated. By processing the physiological data using the corresponding machine learning model, the processing results of the physiological data can be obtained. According to an embodiment of this disclosure, the corresponding machine learning model can identify at least one of sleep stages, respiratory events, and apnea-hypopnea index based on the physiological dataset. For example, sleep stages can include at least one of wakefulness, light sleep, deep sleep, and REM sleep, and respiratory events can include at least obstructive respiratory events.

[0065] According to one embodiment of this disclosure, a corresponding machine learning model can annotate sleep stages and respiratory events based on (e.g., channel-mapped) physiological datasets, and further determine the range of the apnea-hypopnea index (AHI) based on the annotated sleep stages and respiratory durations. The range of AHI can be a core indicator for determining the severity of OSA. According to another embodiment of this disclosure, a corresponding machine learning model can annotate sleep stages and respiratory events based on physiological datasets, and a user, such as a caregiver, can reconfirm the annotated sleep stages and respiratory events. The corresponding machine learning model can determine the range of the apnea-hypopnea index (AHI) based on the confirmed annotated sleep stages and respiratory events. Sleep stages, respiratory events, and the apnea-hypopnea index (AHI) can be intermediate parameters for determining OSA, and can be displayed or otherwise provided to users, such as doctors, so that users can determine the severity of OSA.

[0066] Method 100 can be executed by a processing device such as a user equipment or a computer to achieve the above steps S110-S150.

[0067] Figure 2 This is a flowchart illustrating a method for processing physiological data according to embodiments of the present disclosure. Figure 2As shown, method 200 may include steps S210-S270. Steps in method 200 that are the same as or similar to those in method 100 will not be described in detail.

[0068] In step S220, a machine learning model corresponding to the physiological dataset can be identified. Identifying the machine learning model corresponding to the physiological dataset includes selecting a machine learning model from one or more machine learning models that corresponds to the type of sleep monitoring device.

[0069] According to one embodiment of this disclosure, the detection device may include at least one of a contact sleep detection device and a non-contact sleep detection device. The contact sleep detection device and the non-contact sleep detection device may each correspond to a corresponding machine learning model. (See also...) Figure 3A A detailed description of the non-contact sleep monitoring device. Figure 3A A schematic diagram of a non-contact sleep detection device according to an embodiment of the present disclosure is shown. Figure 3A As shown, the non-contact sleep monitoring device 3000 can be in the form of a pad or a sheet. The non-contact sleep monitoring device 3000 can be equipped with a high-precision sensor (e.g., a 24-bit / 2000 Hz piezoelectric sensor) to capture cardiac impulse signals (BCG), which may include heart rate data, breathing data, snoring data, body movement data, etc. The non-contact sleep monitoring device 3000 can be placed under a pillow to detect physiological data during the user's sleep. In this way, the non-contact sleep monitoring device 3000 can improve the comfort of the user being monitored and the ease of operation. Figure 3A The form of the non-contact sleep detection device 3000 shown may be exemplary, and the non-contact sleep detection device 3000 may also include an energy storage device such as a battery, or have other shapes or sizes, and this disclosure is not limited thereto.

[0070] In step S230, it can be determined whether the first channel of the physiological data in the physiological dataset is the same as the standard channel. For example, it can be determined whether the first channel has a missing channel or other channels compared to the standard channel.

[0071] In step S240, channel mapping can be performed on the physiological data in the physiological dataset in response to the fact that the first channel of the physiological data in the physiological dataset is different from the standard channel.

[0072] Reference Figure 3B and Figure 3C Using physiological data with a first standard channel as an example, an example scenario for step S240 is described, but... Figure 3B and Figure 3C This approach can also be applied to physiological data with a second standard channel.

[0073] Figure 3BThis is an example scenario of processing physiological data according to an embodiment of the present disclosure. The physiological data 3110 with standard channels can be physiological data matched or processed by a first machine learning model. For example... Figure 3B As shown, according to one embodiment of this disclosure, physiological data 3110 with standard channels may include electroencephalogram (EEG) data 3110, eye movement data 3120, electromyography (EMG) data 3130, electrocardiogram (ECG) data 3140, nasopharyngeal airflow data 3150, respiratory effort and pattern data 3160, and blood oxygen saturation data 3170. Physiological dataset 3200 may be detected by a PSG produced by a first manufacturer, and physiological dataset 3300 may be detected by a PSG produced by a second manufacturer. Physiological dataset 3200 may include EEG data 3210, eye movement data 3220, EMG data 3230, ECG data 3240, respiratory effort and pattern data 3260, and blood oxygen saturation data 3270. The physiological dataset 3300 may include electroencephalogram (EEG) data 3310, eye movement data 3320, electrocardiogram (ECG) data 3340, nasopharyngeal airflow data 3350, respiratory effort and pattern data 3360, and blood oxygen saturation data 3370.

[0074] According to one embodiment of this disclosure, the physiological dataset 3200 may have missing channels compared to the physiological data 3110 which has standard channels. For example, due to different configurations from different manufacturers, the PSG produced by the first manufacturer may not detect nasopharyngeal airflow data, thus causing missing channels. Furthermore, Figure 3B The diagram illustrates channel missing in the form of incomplete nasopharyngeal airflow data, but channel missing can exist in other forms. For example, due to different manufacturer configurations, a PSG produced by the first manufacturer may detect nasopharyngeal airflow data using only fewer channels compared to the nasopharyngeal airflow data 3150. Therefore, the physiological dataset 3200 may include nasopharyngeal airflow data, but the nasopharyngeal airflow data may be incomplete (e.g., some channels are missing for multiple channels of the nasopharyngeal airflow data).

[0075] According to one embodiment of this disclosure, the physiological dataset 3300 may have missing channels compared to the physiological data 3110 which has standard channels. For example, due to different configurations from different manufacturers, the PSG produced by the first manufacturer may not detect electromyography data, thus causing missing channels. Furthermore, Figure 3BThe diagram illustrates channel loss in the form of missing all electromyography (EMG) data, but channel loss can exist in other forms. For example, due to different manufacturer configurations, a PSG manufactured by a first manufacturer may detect EMG data using only fewer channels than EMG 3130. Therefore, the physiological dataset 3300 may include EMG data, but the EMG data may be incomplete (e.g., some channels are missing for multiple channels of the EMG data). As another example, a PSG manufactured by a first manufacturer may detect physiological data using the same channels as physiological data 3110 with standard channels. However, due to the detachment of physiological data acquisition components (e.g., electrodes, etc.), the EMG data may have some or all channels missing.

[0076] For channels missing in the first channel compared to the second channel, channel mapping of the physiological data in the physiological dataset may include filling the missing channels with a mask or filling with channels in the first channel that are redundant compared to the standard channels (if any). Furthermore, in response to the presence of missing channels in the first channel of the physiological data in the physiological dataset compared to the standard channels of the physiological data 3110 with standard channels, the weights corresponding to the first channel in the corresponding machine learning model can be increased. That is, the weights of channels (e.g., mapped) corresponding to the channels present in the physiological dataset can be increased in the corresponding machine learning model. In this way, the physiological data processing method according to embodiments of this disclosure can be adapted to various types of detection devices, such as contact sleep detection devices and non-contact sleep detection devices. The physiological data processing method according to embodiments of this disclosure can be adapted to detection devices manufactured by various manufacturers, such as contact sleep detection devices and non-contact sleep detection devices manufactured by different manufacturers. Furthermore, the physiological data processing method according to embodiments of this disclosure can mitigate or eliminate the impact caused by the data acquisition component (e.g., electrodes) detaching or separating from the user (e.g., the user's skin).

[0077] Figure 3C This is an example scenario of processing physiological data according to an embodiment of the present disclosure. The physiological data 3410 with standard channels may be physiological data matched or processed by a first machine learning model. For example... Figure 3CAs shown, according to one embodiment of this disclosure, physiological data 3410 with standard channels may include electroencephalogram (EEG) data 3410, eye movement data 3420, electromyography (EMG) data 3430, electrocardiogram (ECG) data 3440, nasopharyngeal airflow data 3450, respiratory effort and pattern data 3460, and blood oxygen saturation data 3470. Physiological dataset 3500 may be detected by a PSG produced by a third manufacturer, and physiological dataset 3600 may be detected by a PSG produced by a fourth manufacturer. Physiological dataset 3500 may include EEG data 3510, eye movement data 3520, EMG data 3530, ECG data 3540, nasopharyngeal airflow data 3550, respiratory effort and pattern data 3560, and blood oxygen saturation data 3570. The physiological dataset 3600 may include data A 3610 as electroencephalogram (EEG) data, data B 3620 as eye movement data, data C 3630 as electromyography (EMG) data, data D 3640 as electrocardiogram (ECG) data, data E 3650 as nasopharyngeal airflow data, data F 3660 as respiratory effort and pattern data, and data G 3670 as blood oxygen saturation data.

[0078] According to one embodiment of this disclosure, mapping the first channel to the standard channel based on the mapping relationship further includes performing data transformation on the physiological data in the physiological dataset to match it with the physiological data having the standard channel.

[0079] In one embodiment, the encoding method of the electroencephalogram (EEG) data 3410 included in the physiological data 3410 with standard channels can be a first of the following formats: “.cnt”, “.eeg”, “.avg”, “.vhdr”, “.vmrk”, “.mff”, “.raw”, “.bdf”, “.21e”, “.pnt”, “.log”, “.trc”, “.avr”, “.mul”, “.sel”, “.sdt”, “.ent”, and “.pos” (this disclosure is not limited thereto). The EEG data 3510 in the physiological dataset 3500 detected by PSG produced by a third manufacturer can be a second of the following formats different from the first one described above: “.cnt”, “.eeg”, “.avg”, “.vhdr”, “.vmrk”, “.mff”, “.raw”, “.bdf”, “.21e”, “.pnt”, “.log”, “.trc”, “.avr”, “.mul”, “.sel”, “.sdt”, “.ent”, and “.pos”. The encoding method of physiological data in the physiological dataset can be converted to match the corresponding physiological data with standard channels. For example, the encoding method of EEG data 3510 can be converted from the second method to the first method.

[0080] In one embodiment, the sampling frequency of the electromyography (EMG) data 3430 included in the physiological data 3410 with standard channels can be a first sampling frequency among various sampling frequencies. The EMG data 3530 in the physiological dataset 3500 detected by a PSG manufactured by a third manufacturer can be acquired at a second sampling frequency different from the first sampling frequency. The sampling frequency of the physiological data in the physiological dataset can be converted to match the corresponding physiological data with standard channels; for example, the sampling frequency of the EMG data 3530 can be converted from the second sampling frequency to the first sampling frequency.

[0081] In one embodiment, the unit of nasopharyngeal airflow data 3450 included in the physiological data 3410 with standard channels may be a first unit, such as ml / s or liters / s (this disclosure is not limited thereto). The nasopharyngeal airflow data 3550 in the physiological dataset 3500 detected by a PSG manufactured by a third manufacturer may be a second unit, such as ml / s or liters / s, different from the first unit described above. The units of the physiological data in the physiological dataset can be converted to match the corresponding physiological data with standard channels; for example, the units of the nasopharyngeal airflow data 3550 can be converted from the second unit to the first unit.

[0082] In one embodiment, the blood oxygen saturation data 3470 included in the physiological data 3410 with standard channels can be of a first data type, either floating-point or fixed-point (disclosed herein). The blood oxygen saturation data 3570 in the physiological dataset 3500 detected by a PSG manufactured by a third manufacturer can be of a second data type, either floating-point or fixed-point. The data type of the physiological data in the physiological dataset can be converted to match the corresponding physiological data with standard channels; for example, the data type of the blood oxygen saturation data 3570 can be converted from the second data type to the first data type.

[0083] In one embodiment, data A3610, data B3620, data C3630, data D3640, data E3650, data F3660, and data G3670 in the physiological dataset 3600 detected by the PSG produced by a fourth manufacturer can sequentially correspond to electroencephalogram (EEG) data, eye movement (EMG) data, electromyography (EMG) data, electrocardiogram (ECG) data, nasopharyngeal airflow data, respiratory effort and pattern data, and blood oxygen saturation data. The names of the physiological data in the physiological dataset can be converted to match the corresponding physiological data with standard channels. For example, the names of the physiological data in the physiological dataset 3600, "Data A," "Data B," "Data C," "Data D," "Data E," "Data F," and "Data G," can be sequentially converted to "EEG data," "Eye movement (EMG) data," "Electromyography (EMG) data," "Electrocardiogram (ECG) data," "Nasopharyngeal airflow data," "Respiratory effort and pattern data," and "Blood oxygen saturation data."

[0084] For reference Figure 3B and Figure 3C As described, because channel mapping can be performed on the physiological dataset detected by the PSG (e.g., masking missing channels and data transformation of physiological data in the dataset), the method according to embodiments of this disclosure is compatible with different models of PSGs from different manufacturers. Therefore, it avoids the need for medical institutions to repeatedly install different models for processing physiological data on sleep monitoring devices from different manufacturers and to repeatedly perform security authentication processes. Furthermore, due to its strong compatibility, the method according to embodiments of this disclosure avoids extensive training for medical personnel on different models for processing physiological data. Thus, the method provided by embodiments of this disclosure improves its clinical applicability. In addition, the method provided by embodiments of this disclosure not only enhances the diagnostic capabilities of primary healthcare institutions for sleep disorders but also helps to achieve the downward flow of medical resources and improve the balance of regional medical services.

[0085] Similarly, since channel mapping (e.g., masking missing channels and data transformation of physiological data in the dataset) can be performed on the physiological datasets detected by home sleep apnea monitors and non-contact sleep detection devices, the method according to embodiments of this disclosure can obtain physiological data that can be processed by the corresponding machine learning model. In other words, the method according to embodiments of this disclosure is compatible with different models of home sleep apnea monitors and non-contact sleep detection devices from different manufacturers, allowing users to self-test for sleep disorders at home, reducing the burden of cross-regional medical treatment and alleviating the problem of insufficient medical resource allocation. At the same time, due to its strong compatibility, the method according to embodiments of this disclosure can reduce the operational complexity of home sleep disorder testing and improve user compliance with home-based sleep disorder testing.

[0086] In this way, the physiological data processing method according to the embodiments of this disclosure can be adapted to detection devices manufactured by various manufacturers, such as contact sleep detection devices and non-contact sleep detection devices manufactured by different manufacturers.

[0087] Furthermore, abnormal data can be removed from physiological datasets through signal energy threshold detection and morphological analysis. In this way, severely distorted segments of physiological data in physiological datasets (e.g., segments with severely detached physiological data acquisition components) can be identified and removed, ensuring the reliability of subsequent analyses.

[0088] According to embodiments of this disclosure, the data in a physiological dataset can be preprocessed. For example, physiological data from all channels can be filtered, and dimensionality reduction operations such as time-frequency plot calculation can be performed on high-sampling-rate physiological data; however, this disclosure is not limited thereto. Specifically, time-frequency plot calculations can be performed on the physiological data in the physiological dataset. For example, as... Figure 3D As shown, time-frequency plotting can be performed on EEG data to obtain an EEG spectrogram. R-peaks can be extracted from physiological data in a physiological dataset. For example, R-peaks can be extracted from ECG data to obtain the heartbeat signal. The R-peak extraction operation locates the position of the highest positive peak (i.e., the R-wave peak) within each QRS complex (representing ventricular depolarization) in the ECG data. Bandpass filtering can be performed on physiological data in a physiological dataset. For example, bandpass filtering can be performed on raw respiratory data to obtain filtered respiratory data.

[0089] In step S250, the physiological data in the channel-mapped physiological dataset can be input into the corresponding machine learning model for processing. For example, the physiological data in the channel-mapped physiological dataset can be input into the corresponding machine learning model, and the corresponding machine learning model can be used to process the physiological data in the channel-mapped physiological dataset.

[0090] According to one embodiment of this disclosure, a machine learning model corresponding to a physiological dataset can be selected from one or more machine learning models. As previously described, the one or more machine learning models include a first machine learning model corresponding to a contact sleep detection device in a sleep detection device and a second machine learning model corresponding to a non-contact sleep detection device in a sleep detection device. According to an embodiment of this disclosure, the second machine learning model can be obtained from the first machine learning model based on transfer learning.

[0091] For example, one or more machine learning models are trained based on physiological data with standard channels. For instance, a first machine learning model in one or more machine learning models may be trained based on physiological data with a first standard channel, which may include at least information about respiratory and heartbeat signals obtained from a contact sleep monitoring device. Pre-training can be performed based on the physiological data with the first standard channel to obtain the first machine learning model. Based on transfer learning, the first machine learning model is fine-tuned using physiological data with a second standard channel to obtain a second machine learning model, wherein the physiological data with the second standard channel may include at least information about respiratory and heartbeat signals obtained from a non-contact sleep monitoring device.

[0092] In some embodiments, collecting publicly available physiological data with a first standard channel from multiple large-scale sleep studies is considered to ensure diversity and robustness of model training. Physiological data with a first standard channel may include:

[0093] • Sleep Heart Health Study (SHHS): A multicenter cohort study investigating the impact of sleep-disordered breathing on cardiovascular health. It includes nighttime PSG records from over 6,000 participants (primarily middle-aged and older adults).

[0094] • Cleveland Children's Sleep and Health Study

[0095] CCSHS primarily collects data at Cleveland University Hospital Medical Center. This study collected children's sleep data, including sleep duration, sleep quality, symptoms of sleep disorders, and the relationship between these sleep characteristics and children's health.

[0096] • Pediatric Tonsillectomy Study: A randomized clinical trial examining the effects of adenotonsillectomy on children with obstructive sleep apnea. The dataset includes nocturnal PSG recordings from 464 children aged 5–9 years.

[0097] • Home Positive Airway Pressure Study (HomePAP): A study that evaluated home and laboratory diagnostic and treatment methods for sleep apnea and provided nocturnal PSG and portable monitoring data.

[0098] • Multiracial Study of Atherosclerosis (MESA): A cohort study investigating the relationship between sleep characteristics and cardiovascular health, including PSG data from more than 2,000 adults from diverse racial backgrounds.

[0099] • Male Osteoporotic Fracture Study (MrOS): A longitudinal study exploring sleep disorders in older men (≥65 years), with PSG data from more than 3,000 participants.

[0100] • Wisconsin Sleep Cohort (WSC): A longitudinal epidemiological study investigating the natural history of sleep apnea, characterized by collecting overnight PSG data from a cohort of middle-aged adults every four years.

[0101] • Osteoporotic fracture (SOF) study: A large cohort study examining sleep disorders in older women, including PSG and motion recording data.

[0102] • Pediatric adenoidectomy for snoring (PATS): A study evaluating the efficacy of adenoidectomy in children with habitual snoring but without severe obstructive sleep apnea (OSA), including a detailed PSG assessment.

[0103] • Numerous pregnancy outcomes study: Monitoring mothers (NuMoM2B): A study assessing sleep-disordered breathing in pregnant women and its impact on maternal and fetal health, using nighttime PSG recordings from more than 3,000 participants.

[0104] • Human Sleep Project (HSP): This dataset is an expanded compilation of clinical PSG records. Starting with PSG records from approximately 19,000 patients evaluated at Massachusetts General Hospital, the HSP was designed to expand over subsequent years to include data from over 200,000 patients and individuals assessed beyond the clinical setting.

[0105] In this embodiment, the specific type of the first machine learning model is not limited. Any neural network model or combination thereof can be used as the first machine learning model in this embodiment, as long as it can achieve the above-mentioned sleep monitoring function after training.

[0106] In some examples, pre-training based on physiological data with a first standard channel may include: training using information about respiratory and heartbeat signals from the physiological data with a first standard channel as a training set; and testing the ability of the first machine learning model to predict the electrical activity of the brain or heart using information about electroencephalogram (EEG) signals or blood oxygen saturation signals from the physiological data with a first standard channel as a test set.

[0107] In the fine-tuning phase, the pre-trained model is trained on the target domain using physiological data with a second standard channel. The network structure remains unchanged compared to the pre-training phase. However, only the last few layers of the model are unfrozen for training on the physiological data with the second standard channel. This process, often referred to as "fine-tuning" in transfer learning, allows the model to retain knowledge from the source domain while adapting and learning new representations from the target domain.

[0108] This multi-source transfer learning framework yields a model that can perform sleep staging and sleep apnea estimation using only physiological data with a second standard channel as input.

[0109] According to one embodiment of this disclosure, one or more machine learning models can be trained based on corresponding physiological data with standard channels. For example, a first machine learning model can be trained based on previously acquired physiological data with a first standard channel, and a second machine learning model can be trained based on previously acquired physiological data with a second standard channel.

[0110] According to embodiments of this disclosure, during the training phase of a machine learning model, to enhance the robustness of the model to complex low signal-to-noise ratio scenarios, one or more of random white noise, amplitude stretching, and frequency shift can be applied to the corresponding physiological data (e.g., training data) with standard channels to enhance the training data. In this way, problems such as loosening or detachment of physiological data acquisition components (e.g., electrodes) and sampling rate differences in real-world application scenarios can be simulated, thereby improving the robustness of the machine learning model to low-quality signals.

[0111] Reference Figure 3E Describes the process of applying one or more of random white noise, amplitude stretching, and frequency shift to corresponding physiological data with standard channels. Figure 3E This is an example scenario of processing physiological data with a standard channel according to an embodiment of the present disclosure.

[0112] like Figure 3E As shown in the embodiments of this disclosure, in order to simulate unavoidable electrical noise interference in the acquisition system, random white noise can be added to the original training data (e.g., physiological data with standard channels) to generate training data with added white noise. Such noise can be widely present in real-world environments; for example, it can originate from thermal noise from electronic devices, power line interference, and electromyographic activity. By controlling the signal-to-noise ratio level, different levels of noise signals can be generated, thereby generating training data with different noise levels. In this way, machine learning models can learn to extract effective features in noisy backgrounds, thereby enhancing their anti-interference ability and robustness.

[0113] According to embodiments of this disclosure, signal variations caused by improper wearing can be simulated by introducing random amplitude stretching and varying speeds over a time scale. For example, amplitude stretching can be performed on raw training data (e.g., physiological data with standard channels) to generate amplitude-stretched training data. In practical applications, such as electrode patch displacement in ECG monitoring or loosening of the breathing chest and abdomen band, abnormal signal amplitude and distortion of waveform temporal characteristics may occur. Amplitude stretching can simulate signal attenuation or enhancement due to poor contact, as well as temporal shifts caused by sensor slippage or slight changes in physiological rhythm. In this way, the adaptability of machine learning models to fluctuations in the user's electrode wearing quality can be improved.

[0114] According to embodiments of this disclosure, a random frequency offset method can be used to address the problem of inconsistent sampling rates between different detection devices. For example, frequency offset can be performed on the original training data (e.g., physiological data with standard channels) to generate frequency-offset training data. In reality, different types and manufacturers of detection devices can use different sampling frequencies (e.g., sampling frequencies for snoring can have configurations such as 100Hz, 200Hz, and 500Hz). By resampling the original training data (e.g., frequency offset), the training data pattern under different sampling frequencies can be simulated, enabling the model to have feature extraction capabilities across sampling rates and avoiding performance degradation due to sampling rate mismatch. According to one embodiment of this disclosure, physiological data at a single time point of a single channel of physiological data with standard channels can be masked and reconstructed (i.e., random masking) with a predetermined probability, enabling the machine learning model to learn global cross-temporal and cross-channel correlations. For example, as Figure 3FAs shown, mask reconstruction can be performed on individual time points of the EEG data channel, the blood oxygen data channel, the respiration data channel, the heartbeat data channel, and the snoring data channel with predetermined probabilities (shown in dashed boxes). Mask reconstruction (i.e., temporal masking) can be performed on physiological data at the same time points of all channels with standard channels, enhancing the machine learning model's ability to learn the temporal relationships of physiological signals. For example, mask reconstruction can be performed on the same time points of the EEG, blood oxygen, respiration, heartbeat, and snoring data channels (shown in dashed boxes). Mask reconstruction (i.e., channel masking) can be performed on physiological data for the complete time of predetermined channels with standard channels, enhancing the machine learning model's ability to learn the relationships between physiological signal channels. For example, mask reconstruction can be performed on the complete time of physiological data for the blood oxygen and snoring data channels (shown in dashed boxes). It is possible to perform masked reconstruction (i.e., local masking) on ​​physiological data of adjacent channels with standard channels for a predetermined time period, enabling machine learning models to enhance their ability to detect strong correlations between temporal sequences and channels within a local neighborhood. For example, it is possible to perform masked reconstruction on physiological data of blood oxygenation and respiratory data channels for a predetermined time period (shown in the dashed box).

[0115] The above methods are compatible with various models of PSG devices, home sleep apnea monitors, and non-contact sleep testing devices. These methods not only provide auxiliary interpretation in hospitals to improve report generation efficiency through automatic annotation technology, but can also be applied in home settings to mitigate the interference of the hospital first-night effect. Because they assist in identifying parameters such as sleep stages, respiratory events, and apnea-hypopnea index, they reduce reliance on user expertise, allowing primary care physicians to obtain these reliable intermediate parameters through standardized data analysis. Furthermore, when data received from the testing device does not conform to the standards of machine learning models, channel mapping can be used to process the data, thereby improving the accuracy of the processing results. This not only enhances the diagnostic capabilities of primary healthcare institutions for sleep disorders but also helps to achieve the downward flow of medical resources and improve the balance of regional medical services.

[0116] In step S260, in response to the fact that the first channel of the physiological data in the physiological dataset is the same as the standard channel, channel mapping of the physiological data in the physiological dataset may not be performed.

[0117] In step S270, physiological data from the unmapped physiological dataset can be input into the corresponding machine learning model for processing.

[0118] Figure 4 This is a schematic diagram of an apparatus for processing physiological data according to an embodiment of the present disclosure.

[0119] like Figure 4 As shown, the device 400 for processing physiological data may include a receiving module 410, an identification module 420, a mapping module 430, and a processing module 440.

[0120] The receiving module 410 can be configured to receive physiological datasets for sleep monitoring from the detection device.

[0121] The identification module 420 can be configured to identify a machine learning model corresponding to the physiological dataset, wherein the physiological dataset has a first channel, the first channel is related to the data acquisition path of the detection device, and the corresponding machine learning model is configured to process physiological data with a standard channel.

[0122] The mapping module 430 can be configured to determine the mapping relationship between the first channel and the standard channel, and map the first channel to the standard channel based on the mapping relationship.

[0123] The processing module 440 can be configured to input physiological data from the mapped physiological dataset into the corresponding machine learning model for processing.

[0124] According to another aspect of this disclosure, an apparatus for processing physiological data is also provided. Figure 5 A schematic diagram of a device 500 for processing physiological data according to an embodiment of the present disclosure is shown.

[0125] like Figure 5 As shown, the device 500 for processing physiological data may include one or more processors 510 and one or more memories 520. The memories 520 store computer-readable code that, when executed by the one or more processors 510, can perform the method for processing physiological data as described above.

[0126] The processor in the embodiments of this disclosure can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 architecture or an ARM architecture.

[0127] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0128] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 6 The architecture of the computing device 600 shown is used for implementation. For example... Figure 6 As shown, the computing device 600 may include a bus 610, one or more CPUs 620, a read-only memory (ROM) 630, a random access memory (RAM) 640, a communication port 650 connected to a network, an input / output component 660, a hard disk 670, etc. The storage devices in the computing device 600, such as the ROM 670 or the hard disk 670, may store various data or files used for processing and / or communication of the methods for processing physiological data provided in this disclosure, as well as program instructions executed by the CPU. The computing device 600 may also include a user interface 680. Of course, Figure 6 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 6 One or more components in the computing device shown.

[0129] According to another aspect of this disclosure, a computer-readable storage medium is also provided. The computer storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, a method for processing physiological data according to embodiments of this disclosure, as described with reference to the above-drawn figures, can be performed. The computer-readable storage medium in the embodiments of this disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0130] Embodiments of this disclosure also provide a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a method for processing physiological data according to embodiments of this disclosure.

[0131] Embodiments of this disclosure provide a method, apparatus, device, and computer-readable storage medium for processing physiological data.

[0132] The methods, apparatus, devices, and storage media for processing physiological data for monitoring sleep provided in the embodiments of this disclosure are adaptable to a variety of physiological data detection devices, reducing the cost of collecting and analyzing physiological data. Furthermore, the methods according to this disclosure promote the use of low-cost, easy-to-use, and user-friendly (e.g., non-contact) physiological data detection devices, reducing medical costs for users and alleviating the demands on the diagnostic and treatment capabilities of primary healthcare institutions.

[0133] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing at least one executable instruction for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0134] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0135] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations can be made to these embodiments or their features without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A method for processing physiological data used in sleep monitoring, comprising: Receive physiological data sets for sleep monitoring from the detection device; Identify a machine learning model corresponding to the physiological dataset, wherein the physiological dataset has a first channel, the first channel is related to the data acquisition path of the detection device, and the corresponding machine learning model is configured to process physiological data with a standard channel; Determine the mapping relationship between the first channel and the standard channel, and map the first channel to the standard channel based on the mapping relationship; and The physiological data in the mapped physiological dataset is input into the corresponding machine learning model for processing.

2. The method according to claim 1, wherein, Mapping the first channel to the standard channel based on the mapping relationship includes: Determine whether the first channel has any missing channels compared to the standard channel. In response to the first channel having a missing channel compared to the standard channel, the missing channel is filled.

3. The method according to claim 2, wherein, The process of filling the missing channels includes: Determine whether the first channel has any other channels besides the standard channel compared to the standard channel. In response to the existence of the other channels, the missing channels are filled using the other channels.

4. The method according to claim 2, wherein, The process of filling the missing channels includes: Determine whether the first channel has any other channels besides the standard channel compared to the standard channel. In response to the absence of the other channels, the missing channels are filled with a mask.

5. The method according to claim 1, wherein, Mapping the first channel to the standard channel based on the mapping relationship includes: Determine whether the first channel has a missing channel compared to the standard channel, and determine whether the first channel has other channels besides the standard channel compared to the standard channel. In response to the first channel not having a missing channel compared to the standard channel and the existence of the other channels, the other channels are discarded.

6. The method according to claim 1, wherein, The process of inputting physiological data from the mapped physiological dataset into the corresponding machine learning model for processing also includes: Select a machine learning model from one or more that corresponds to the physiological dataset. The physiological data from the mapped physiological dataset is input into the corresponding machine learning model, and The corresponding machine learning model is used to process the physiological data in the mapped physiological dataset.

7. The method according to claim 6, wherein, The type of detection device includes a contact sleep detection device, and the selection of a machine learning model corresponding to the physiological dataset includes selecting a first machine learning model corresponding to the contact sleep detection device from one or more machine learning models, wherein the first machine learning model processes at least electrocardiogram data, chest strap data, and abdominal strap data.

8. The method according to claim 3, wherein, The first machine learning model also processes at least one or more of the following: electroencephalogram (EEG) data, eye movement data, electromyography (EMG) data, airflow channel data, nasal pressure channel data, transcutaneous oxygen saturation channel data, and snoring channel data.

9. The method according to claim 7, wherein, The contact-type sleep monitoring device includes at least one of a polysomnography monitor and a home sleep apnea monitor.

10. The method according to claim 6, wherein, The type of detection device includes a non-contact sleep detection device, and wherein selecting a machine learning model corresponding to the physiological dataset includes selecting a second machine learning model corresponding to the non-contact sleep detection device from one or more machine learning models, the second machine learning model processing at least heart rate data and breathing data.

11. The method according to claim 10, wherein, The second machine learning model also processes at least one of the snoring data and body movement data.

12. The method according to claim 6, wherein, The one or more machine learning models are trained based on physiological data with standard channels.

13. The method according to claim 12, wherein, The physiological data with standard channels are subjected to one or more of random white noise, amplitude stretching, and frequency shift.

14. The method according to claim 12, wherein, The physiological data with standard channels are subjected to one or more of the following operations: Physiological data from a single time point in a single channel of physiological data with a predetermined probability are reconstructed by masking. Physiological data at the same time point from all channels with standard channels are reconstructed using a mask; Masking and reconstructing physiological data for the entire time of a predetermined channel with standard channels; as well as Physiological data from adjacent channels with standard channels for a predetermined time period are reconstructed using a mask.

15. The method according to claim 1, wherein, Mapping the first channel to the standard channel based on the mapping relationship further includes: Transform the names of physiological data in the physiological dataset. The encoding method of physiological data in the physiological dataset is converted. The sampling rate of the physiological data in the physiological dataset is converted. Convert the units of physiological data in the physiological dataset, and Convert the data type of physiological data in the physiological dataset.

16. The method of claim 1, further comprising, in response to the presence of a missing channel in the first channel of the physiological data in the physiological dataset compared to the standard channel, increasing the weight in the corresponding machine learning model corresponding to the mapped first channel.

17. The method according to claim 1, further comprising: Abnormal data are removed from the physiological dataset by signal energy threshold detection and morphological analysis.

18. The method according to claim 6, wherein, The one or more machine learning models include a first machine learning model corresponding to a contact sleep detection device in the sleep detection device and a second machine learning model corresponding to a non-contact sleep detection device in the sleep detection device, and The second machine learning model is obtained from the first machine learning model based on transfer learning.

19. The method according to claim 1, wherein, The physiological data from the mapped physiological dataset is input into the corresponding machine learning model for processing, including: The corresponding machine learning model identifies at least one of sleep stages, respiratory events, and apnea-hypopnea index based on the mapped physiological dataset.

20. The method according to claim 19, wherein, The sleep stages include at least one of wakefulness, light sleep, deep sleep, and REM sleep, and The respiratory events include at least obstructive respiratory events.

21. The method according to claim 1, further comprising preprocessing the physiological data in the physiological dataset, wherein, The preprocessing includes one or more of the following: Time-frequency plot calculations are performed on the physiological data in the aforementioned physiological dataset; Extract the R peak from the physiological data in the physiological dataset; Bandpass filtering is applied to the physiological data in the physiological dataset.

22. An apparatus for processing physiological data for sleep monitoring, the apparatus comprising: The receiving module is configured to receive physiological datasets for sleep monitoring from the detection device; The identification module is configured to identify a machine learning model corresponding to the physiological dataset, wherein the physiological dataset has a first channel, the first channel is related to the data acquisition path of the detection device, and the corresponding machine learning model is configured to process physiological data with a standard channel. A mapping module is configured to determine the mapping relationship between the first channel and the standard channel, and map the first channel to the standard channel based on the mapping relationship; and The processing module is configured to input physiological data from the mapped physiological dataset into the corresponding machine learning model for processing.

23. A device for processing physiological data, comprising: One or more processors; as well as One or more memories storing a computer-executable program, which, when executed by the processor, performs the method of any one of claims 1-21.

24. A computer program product comprising a computer program or instructions, wherein, The computer program or instructions, when executed by a processor, implement the method as described in any one of claims 1-21.

25. A computer-readable storage medium having stored thereon computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-21.

Images