Sleep snoring sound detection and adaptive intervention system based on wrist multi-modal data

Abstract

The present application belongs to the technical field of sleep health monitoring, and relates to a sleep snoring sound detection and adaptive intervention system based on wrist multi-modal data, which comprises a wrist master device and a neck slave device. The former comprises a multi-modal sensing unit for collecting multi-modal physiological data of a user; an audio acquisition unit for collecting audio data; a pre-trigger wake-up module for determining whether the multi-pre-trigger condition is met; a snoring sound source determination module for comprehensively determining whether the user snores; a snoring sound state grading module for calculating an apnea hypopnea index and determining the snoring disorder event level of the user; and an adaptive grading intervention module for generating an intervention instruction corresponding to the snoring disorder event level. The latter comprises a micro piezoelectric vibration sensor for collecting throat vibration characteristics and a micro myoelectric electrode for collecting throat myoelectric characteristics. The present application can realize accurate identification of the snoring sound source, grading evaluation of OSAHS risk, and adaptive matching of intervention strategies.

Description

Technical Field

[0001] This invention belongs to the field of sleep health monitoring technology, specifically relating to a sleep snoring detection and adaptive intervention system based on wrist multimodal data. Background Technology

[0002] Obstructive sleep apnea-hypopnea syndrome (OSAHS) is a common sleep-disordered breathing disorder characterized by recurrent upper airway collapse or obstruction during sleep, leading to apnea or hypopnea events, causing intermittent hypoxemia and sleep structure disturbances. The diagnostic criterion is an apnea-hypopnea index (AHI) ≥5 times / h.

[0003] OSAHS has a high incidence rate in the population, especially among middle-aged and elderly people, and poses a significant threat, leading to various complications such as hypertension, coronary heart disease, stroke, and diabetes. In severe cases, it can even cause sudden death. However, due to insufficient public awareness and limited diagnostic methods, the detection rate of OSAHS is extremely low, and a large number of patients remain undiagnosed and untreated for a long time. Currently, the clinical diagnosis of OSAHS relies on polysomnography (PSG). However, PSG equipment is expensive and complex to operate, requiring patients to have dozens of wires attached to their bodies for overnight monitoring in a hospital sleep center. This monitoring method not only consumes a lot of medical resources, but also easily produces the "first-night effect" due to environmental changes and wire constraints, resulting in monitoring data that cannot accurately reflect the patient's sleep state in their natural home environment.

[0004] With the widespread adoption of smart wearable devices, snoring detection and sleep apnea monitoring products based on wrist-worn devices (smartwatches / bands) and ear-worn devices have emerged on the market. However, these existing technologies have the following drawbacks: For wrist-worn devices (watches / bands), firstly, the microphone is too far from the sound source to distinguish between the user's own snoring, the snoring of someone in the same room, and ambient noise; secondly, intervention relying solely on wrist vibration lacks targeted stimulation, and the vibration intensity is difficult to control precisely, easily waking the user and resulting in insufficient intervention effectiveness; thirdly, they lack OSAHS classification capabilities, only providing vague risk warnings, which lack clinical reference value; and fourthly, some devices have microphones that are always on, leading to short battery life and the risk of privacy leaks.

[0005] For ear-worn devices, firstly, they have a strong foreign body sensation, causing significant pain when the ear canal is compressed while sleeping on one's side, making long-term wear unacceptable to users; secondly, the microphone is always on to collect audio, posing a high risk of missed recordings and raising significant concerns about user privacy.

[0006] In summary, how to fully utilize multimodal physiological sensing information from the wrist, along with optional neck vibration and electromyography sensing information, to accurately identify the source of snoring, classify and assess OSAHS risks, and adaptively match intervention strategies under dynamic changes in the user's wearing status (with or without the neck patch) has become a crucial technical problem that urgently needs to be solved in the field of smart wearable sleep health monitoring. Summary of the Invention

[0007] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a sleep snoring detection and adaptive intervention system based on wrist multimodal data. This system makes full use of wrist multimodal physiological sensing information and optional neck vibration and electromyography sensing information to accurately identify the source of snoring, classify and assess OSAHS risk, and adaptively match intervention strategies under dynamic changes in the user's wearing status (with or without neck patch).

[0008] This invention provides a sleep snoring detection and adaptive intervention system based on wrist multimodal data, including a wrist-based main device: The wrist-mounted main device includes a multimodal sensing unit, an audio acquisition unit, and a core control unit; The multimodal sensing unit is used to collect the user's multimodal physiological data, and the audio acquisition unit is used to collect audio data after wake-up. The core control unit includes: The pre-trigger wake-up module is used to determine whether multiple pre-trigger conditions are met based on the multimodal physiological data, and wake up the audio acquisition unit when the multiple pre-trigger conditions are met; The snoring source determination module is used to comprehensively determine whether the user is snoring based on the audio data and the multimodal physiological data. The snoring status grading module is used to determine the apnea-hypopnea index based on the audio data and the multimodal physiological data after the user starts snoring, and to determine the level of the user's snoring symptom. The adaptive grading intervention module is used to generate corresponding intervention instructions based on the snoring symptom event level, and call the intervention processing module to execute the intervention instructions.

[0009] In one embodiment of the present invention, a neck support device is also included: The neck slave device includes a miniature piezoelectric vibration sensor and a miniature electromyographic electrode. The neck slave device is fixed to the user's throat on the side of the neck. The wrist master device and the neck slave device are connected to each other through built-in wireless communication units.

[0010] In one embodiment of the present invention, the multimodal sensing unit includes a six-axis inertial measurement module, a blood oxygen pulse sensing module, and an electrocardiogram sensing module; The six-axis inertial measurement module is used to collect body motion and posture features, the blood oxygen pulse sensing module is used to collect blood oxygen features, the electrocardiogram sensing module is used to collect heart rate features, the miniature piezoelectric vibration sensor is used to collect laryngeal vibration features, and the miniature electromyography electrode is used to collect laryngeal electromyography features.

[0011] In one embodiment of the present invention, when the pre-triggered wake-up module performs the step of determining whether multiple pre-triggered conditions are met based on the multimodal physiological data, it includes: The multiple pre-triggering conditions include sleep state conditions, body movement state conditions, and feature triggering conditions; Sleep staging results are generated based on the body movement characteristics, posture characteristics, and blood oxygenation characteristics. If the sleep staging result is a preset sleep stage, then the sleep state conditions are met. The amplitude of the user's body movement is detected based on the body movement characteristics. If the amplitude of the body movement is less than a preset body movement threshold, the body movement state condition is met. When connecting a neck-mounted device, if the throat vibration characteristics match the preset snoring characteristic frequency band, the characteristic triggering condition is met. When the neck-worn device is not connected, the rate of decrease in blood oxygen is calculated based on the blood oxygen characteristics, and the sleeping posture is determined based on the posture characteristics. If the rate of decrease in blood oxygen is greater than a preset first blood oxygen threshold and the posture characteristics are a supine sleeping posture, then the feature triggering condition is met.

[0012] In one embodiment of the present invention, when the snoring source determination module performs the step of comprehensively determining whether the user is snoring based on the audio data and the multimodal physiological data, it includes: Calculate the snoring similarity between the audio data and the user-pre-acquired snoring template; Calculate the vibration similarity between the throat vibration characteristics and the user-pre-acquired throat vibration template; Calculate the blood oxygen similarity between the blood oxygen features and the pre-acquired snoring blood oxygen template; Calculate the laryngeal electromyographic similarity between the laryngeal electromyographic features and the pre-acquired snoring laryngeal electromyographic template; The snoring similarity, vibration similarity, blood oxygenation similarity, and laryngeal electromyography similarity are dynamically weighted and calculated to obtain the individual's snoring similarity. If the similarity of the user's snoring to a preset similarity threshold is greater than the threshold, the user is identified as snoring; otherwise, the sound is identified as environmental or roommate noise, and the audio acquisition unit is turned off.

[0013] In one embodiment of the present invention, when the snoring state grading module performs the calculation of the apnea-hypopnea index based on the audio data and the multimodal physiological data, it includes: Within a preset sliding detection window, it is determined whether a sleep apnea event is met based on the audio data, the blood oxygen characteristics, and the heart rate characteristics. In the sliding detection window, it is determined whether a hypoventilation event is met based on the laryngeal vibration characteristics and the blood oxygenation characteristics. The apnea and hypopnea events of all sliding detection windows within a preset sleep time range are statistically calculated to obtain the apnea-hypopnea index.

[0014] In one embodiment of the present invention, when the snoring state grading module performs the calculation of the apnea-hypopnea index based on the audio data and the multimodal physiological data, it includes: Within a preset sliding detection window, the snoring interruption time is determined based on the audio data and / or the laryngeal vibration characteristics, the blood oxygen decrease is calculated based on the blood oxygen characteristics, and the heart rate increase is calculated based on the heart rate characteristics. If the snoring interruption time under the sliding detection window is greater than the preset interruption threshold, the blood oxygen drop is greater than the preset second blood oxygen threshold, and the heart rate rise is greater than the preset heart rate threshold, then the sleep apnea event is determined to be satisfied. In the sliding detection window, the decrease in snoring intensity is determined based on the laryngeal vibration characteristics, the duration of snoring is determined based on the laryngeal vibration characteristics and / or audio data, and the decrease in blood oxygen is calculated based on the blood oxygen characteristics. If the decrease in snoring intensity under the sliding detection window is greater than a preset intensity threshold, the duration of snoring is greater than a preset duration threshold, and the decrease in blood oxygen is greater than a preset third blood oxygen threshold, then the hypoventilation event is determined to be met. The total number of events was obtained by counting apnea and hypopnea events in all sliding detection windows within a preset sleep time range. The apnea-hypopnea index is calculated by comparing the total number of events with the total duration of the sleep time range.

[0015] In one embodiment of the present invention, the adaptive graded intervention module further includes: when a hypoventilation event is determined to be met, generating a first intensity value electrical stimulation command and / or a first intensity value vibration command as an intervention command.

[0016] In one embodiment of the present invention, when the adaptive grading intervention module executes the intervention instruction generated according to the snoring symptom event level, it includes: When the snoring symptom level is mild, a second-intensity electrical stimulation command and / or a second-intensity vibration command are generated as intervention commands. Wherein, the first intensity value in the first intensity value electrical stimulation command is greater than the second intensity value in the second intensity value electrical stimulation command, and the first intensity value in the first intensity value vibration command is greater than the second intensity value in the second intensity value vibration command; When the snoring symptom is classified as moderate or severe, a graded early warning instruction and a linkage device control instruction are generated.

[0017] In one embodiment of the present invention, when the adaptive hierarchical intervention module executes the intervention instruction by calling the intervention processing module, it includes: The intervention module includes a wrist vibration unit located in the main device at the wrist, a subthreshold electrical stimulation unit located in the slave device at the neck, a user terminal display unit, and an external ventilator; The wrist vibration unit is used to execute the first intensity value vibration command and the second intensity value vibration command; The subthreshold electrical stimulation unit is used to execute the first intensity value electrical stimulation command and the second intensity value electrical stimulation command; The user terminal display unit is used to execute the graded early warning command; The external ventilator is wirelessly connected to the wrist-mounted main device and is used to execute control commands for the linked device.

[0018] The beneficial effects of this invention are as follows: Through a multi-condition pre-triggering mechanism, this invention enables on-demand microphone activation in the physical circuit, significantly extending the device's battery life and eliminating user anxiety about privacy leaks; by dynamically weighting and matching multi-dimensional physiological characteristics, a rigorous snoring verification process is constructed, significantly improving the accuracy of personal snoring recognition in complex environments and overcoming the shortcomings of traditional single acoustic detection which is susceptible to interference; by introducing a sliding detection window and accurately mapping clinical diagnostic standards, localized high-precision estimation of AHI is achieved; based on objective symptom grading results, a step-by-step adaptive intervention is implemented, from non-invasive subthreshold electrical stimulation to linkage with a home ventilator, effectively preventing airway collapse and deterioration caused by snoring while greatly reducing damage to normal sleep structure and improving long-term user compliance. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 This is a functional block diagram of a wrist main device provided in one embodiment of the present invention; Figure 2 This is a functional block diagram of a neck slave device provided in one embodiment of the present invention; Figure 3 This is a functional block diagram of a sleep snoring detection and adaptive intervention system based on wrist multimodal data provided in one embodiment of the present invention; The system includes: a wrist main device (100), a multimodal sensing unit (110), a six-axis inertial measurement module (111), a blood oxygen pulse sensing module (112), an electrocardiogram sensing module (113), an audio acquisition unit (120), a core control unit (130), a pre-triggered wake-up module (131), a snoring source determination module (132), a snoring state grading module (133), an adaptive grading intervention module (134), a wrist vibration unit (140), a neck slave device (200), a miniature piezoelectric vibration sensor (210), a miniature electromyographic electrode (220), a subthreshold electrical stimulation unit (230), a user terminal display unit (300), and an external ventilator (400). Detailed Implementation

[0021] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0022] This invention provides a sleep snoring detection and adaptive intervention system based on wrist multimodal data. This application is primarily used in the field of sleep health monitoring and intervention, particularly for the routine screening and non-invasive intervention of obstructive sleep apnea-hypopnea syndrome (OSAHS). OSAHS is a chronic disease characterized by apnea or hypopnea during sleep due to partial or complete collapse of the upper airway, leading to intermittent hypoxemia and disruption of sleep structure. It not only causes daytime sleepiness and cognitive decline but is also an independent risk factor for cardiovascular and cerebrovascular diseases such as hypertension, coronary heart disease, and stroke.

[0023] Although various smartwatches, wristbands, and anti-snoring devices with sleep monitoring functions already exist on the market, these technologies have revealed numerous irreconcilable technical contradictions and defects in practical applications. Firstly, the misidentification rate of snoring sources is extremely high. Traditional wrist-worn devices typically rely solely on built-in microphones for acoustic data collection. Because the wrist is far from the sound source (mouth and nose), and the acoustic environment of a bedroom is complex, the device easily misidentifies the snoring of a roommate, the sound of an air conditioner, or ambient white noise from outside the window as the wearer's own snoring. This single-dimensional acoustic judgment cannot achieve precise spatial localization of the sound source at the physical level, resulting in a persistently high false positive rate in the monitoring data.

[0024] Secondly, there is an irreconcilable contradiction between microphone power management and privacy protection in existing technologies. To ensure no snoring event is missed, some devices employ a strategy of keeping the microphone on all night. This high-frequency audio sampling and continuous analog-to-digital conversion cause the tiny battery capacity of smart wearable devices to be depleted in a short time, often resulting in a battery life of less than 24 hours, severely impacting the user experience. More critically, continuous overnight audio recording easily triggers strong privacy concerns among users about their private conversations in the bedroom being overheard, leading many users to refuse to use such features.

[0025] Furthermore, existing intervention methods are often too harsh, easily leading to disturbances and sleep deprivation. When a device detects snoring, it typically outputs a fixed intensity of mechanical vibration or electric shock to force the user to change their sleeping position. However, this "one-size-fits-all" intervention, which does not differentiate between different levels of pathological severity, often directly wakes users from deep sleep, disrupting their normal sleep cycle. In the long run, this not only fails to effectively treat OSAHS but also exacerbates daytime fatigue, resulting in extremely poor user compliance.

[0026] Finally, the vast majority of consumer-grade devices lack the ability to classify symptoms with clinical reference value. They typically only provide a binary judgment of "whether or not snoring occurs," or offer a vague "sleep quality score." In clinical medicine, the severity of OSAHS is objectively quantified by the Apnea-Hypopnea Index (AHI). Lacking the simultaneous support of multi-dimensional physiological data (such as blood oxygenation, heart rate, body movement, and airflow obstruction status), consumer-grade devices are fundamentally unable to accurately estimate AHI, resulting in data silos that cannot provide any valuable reference for doctors' diagnoses or the parameter adjustment of home ventilators.

[0027] To address the aforementioned technical challenges, this invention provides a sleep snoring detection and adaptive intervention system based on wrist multimodal data. It abandons the isolated judgment of a single sensor, constructing a "master-slave collaborative" hardware architecture and a "multimodal fusion" algorithm. By introducing a multi-condition joint pre-triggered wake-up mechanism, it achieves precise on-demand activation of the high-power audio module, balancing long battery life and strong privacy. Through dynamic weighted matching of multi-dimensional features such as audio, vibration, blood oxygen, and electromyography, and cross-verification using multi-source data, it improves the problem of misjudging the source of snoring. By introducing a sliding detection window, it accurately maps the clinical medical criteria for apnea and hypopnea, achieving high-precision estimation of AHI. Finally, based on objective disease grading results, it implements a step-by-step adaptive intervention, from imperceptible subthreshold stimulation to linkage with a home ventilator, ensuring airway patency while keeping the awakening rate to a minimum.

[0028] See Figures 1 to 3 As shown in the figure, the sleep snoring detection and adaptive intervention system based on wrist multimodal data provided by the present invention includes: a wrist master device 100. In order to further improve the detection accuracy and intervention effect and solve the problem of limited monitoring dimensions of a single wrist device, the system may also optionally include a neck slave device 200.

[0029] The neck support device 200 is fixed to the user's neck and throat (e.g., a reusable neck patch is attached approximately two centimeters above the Adam's apple). It integrates a miniature piezoelectric vibration sensor 210, a miniature electromyographic electrode 220, and a subthreshold electrical stimulation unit 230. The neck support device 200 is an optional accessory; users can choose whether to wear it based on their needs. When worn, it provides higher detection accuracy and a gentler intervention effect. When not worn, the system can still operate independently, flexibly adapting to different user preferences.

[0030] The overall thickness of the device 200 is strictly controlled to be within an extremely thin range, preferably within 0.5 mm, so that when users sleep on their side, lie on their stomach, or frequently turn over, there will be no obvious pressure, friction, or pain on the skin and subcutaneous tissue, thereby greatly improving the user's compliance with long-term home wear.

[0031] Specifically, the miniature piezoelectric vibration sensor 210 in the neck device 200 is attached to the skin of the throat to collect the vibration characteristics of the throat. When snoring occurs, the airflow impacts the relaxed soft palate, uvula and pharyngeal wall, and the resulting mechanical waves are directly transmitted to the surface of the neck skin. The sensor uses the piezoelectric effect to convert this tiny mechanical deformation into a high-fidelity electrical signal in real time.

[0032] The miniature electromyography electrode 220 forms an electrical connection with the skin surface to collect the electromyographic characteristics of the larynx and monitor the surface electromyographic activity potential of key muscles in the pharynx (especially the genioglossus muscle) in real time. If the electromyographic activity of the genioglossus muscle is significantly weakened during sleep, it often indicates that the airway is about to collapse obstructively.

[0033] To enable data flow and logical collaboration between master and slave devices, the wrist master device 100 and the neck slave device 200 communicate with each other through their built-in wireless communication unit (located on the wrist master device 100 side) and low-power Bluetooth communication unit (located on the neck slave device 200 side), respectively, to achieve data interaction and command transmission.

[0034] In this embodiment of the invention, the wrist main device 100 includes a multimodal sensing unit 110, an audio acquisition unit 120, a wrist vibration unit 140, and a core control unit 130. The multimodal sensing unit 110 is used to collect the user's multimodal physiological data. Sleep is a complex physiological process involving multiple systems such as the nervous, respiratory, and cardiovascular systems. Single-dimensional physiological data often only reflects the surface and is easily affected by environmental noise or individual differences. By integrating multiple sensors, the multimodal sensing unit 110 can characterize the user's physiological activities during sleep from multiple dimensions such as kinematics and electrophysiology.

[0035] Furthermore, the multimodal sensing unit 110 specifically includes a six-axis inertial measurement module 111, a blood oxygen pulse sensing module 112, and an electrocardiogram sensing module 113; the six-axis inertial measurement module 111 is used to collect body motion features and attitude features, the blood oxygen pulse sensing module 112 is used to collect blood oxygen features, and the electrocardiogram sensing module 113 is used to collect heart rate features.

[0036] The six-axis inertial measurement module 111 is typically composed of a three-axis accelerometer and a three-axis gyroscope. It is used to collect the user's three-dimensional spatial acceleration and angular velocity changes at high frequency, thereby extracting body motion and posture features. The system utilizes the projection distribution of the gravitational acceleration vector on the three axes, combined with the angular velocity integral of the gyroscope, to calculate the user's current sleeping posture (such as supine, left lateral, right lateral, prone, etc.). The blood oxygen pulse sensing module 112 is preferably a PPG photoplethysmography module. Airway obstruction leads to a decrease in alveolar ventilation, which in turn causes a decrease in arterial blood oxygen partial pressure. Therefore, the stepwise or V-shaped / U-shaped decrease in blood oxygen can be used to determine apnea and hypoventilation events. The electrocardiogram sensing module 113 is preferably a single-lead ECG module. When apnea occurs and hypoxia occurs, the brain will generate micro-awakening, triggering abnormal excitation of the sympathetic nervous system, which manifests as a compensatory spike in heart rate.

[0037] The audio acquisition unit 120 (e.g., a high-sensitivity miniature microphone) is used to acquire audio data after being woken up. Unlike microphones in traditional devices that are always in a listening state, the audio acquisition unit 120 in this invention is in a deep sleep mode by default and is only activated after receiving a specific wake-up signal from the core control unit 130. Specifically, it can acquire audio data of a preset short duration (e.g., 5 to 10 seconds).

[0038] The core control unit 130 is the system's computing and control center. It integrates a pre-trigger wake-up module 131, a snoring source determination module 132, a snoring state grading module 133, and an adaptive grading intervention module 134. These four modules are executed sequentially on the time axis, forming a complete closed-loop control flow of "perception-analysis-diagnosis-execution".

[0039] Specifically, the pre-triggered wake-up module 131 is used to determine whether multiple pre-triggered conditions are met based on the multimodal physiological data, and wakes up the audio acquisition unit 120 when the multiple pre-triggered conditions are met; the microphone will only be triggered when the multimodal physiological data shows a specific combination of features that are highly suspected to be snoring or sleep apnea. Through this funnel-shaped screening mechanism with multiple conditions combined, the system can minimize the number of invalid microphone activations.

[0040] The snoring source determination module 132 is used to comprehensively determine whether the user is snoring based on the audio data and the multimodal physiological data. The captured audio data is strictly time-stamped with the multimodal physiological data (such as small fluctuations in blood oxygen, body movements in a specific posture, etc.) within the same time window. Through mutual verification and cross-validation of multi-source data, this module can effectively remove the interference of environmental white noise, air conditioning noise and snoring from roommates, thereby significantly improving the accuracy of snoring determination.

[0041] The snoring status grading module 133 is used to determine the user's snoring status, calculate the apnea-hypopnea index (AHI) based on the audio data and multimodal physiological data, and determine the user's snoring symptom level. Assuming the user is indeed snoring, this module further delves into pathological characteristics, identifying each instance of complete airway obstruction (apnea) and partial obstruction (hypopnea), and then calculates the AHI, the gold standard for clinical diagnosis of OSAHS. Based on the AHI value, this module classifies the user's current state into different symptom levels such as normal, mild, moderate, and severe.

[0042] The adaptive grading intervention module 134 generates corresponding intervention instructions based on the snoring symptom event level and calls the intervention processing module to execute the intervention instructions. This module dynamically and adaptively matches intervention strategies according to the severity of the symptom. For mild cases, it prioritizes outputting weak, imperceptible stimuli; for severe cases, it upgrades intervention methods or connects to external professional medical equipment. This tiered intervention logic maximizes airway patency while significantly reducing disruption to normal sleep structure.

[0043] The intervention module includes a wrist vibration unit 140 located in the wrist master device 100, a subthreshold electrical stimulation unit 230 located in the neck slave device 200, a user terminal display unit 300, and an external ventilator 400.

[0044] In this embodiment of the invention, during the pre-trigger wake-up phase, the system defaults to shutting down the audio acquisition unit 120 to save power and protect user privacy, and only wakes up the audio acquisition unit 120 for short-term acquisition when all multiple pre-trigger conditions are met.

[0045] Specifically, the multiple pre-triggered conditions include sleep state conditions, body movement state conditions, and feature triggering conditions.

[0046] Sleep is typically divided into wakefulness, light sleep, deep sleep, and REM sleep. Severe snoring and sleep apnea mostly occur during deep sleep and REM sleep. Therefore, the method for determining sleep state conditions is as follows: a sleep staging result is generated based on the body movement characteristics, posture characteristics, and blood oxygenation characteristics. If the sleep staging result is a preset sleep stage (such as deep sleep and / or REM sleep), then the sleep state conditions are met.

[0047] If the amplitude of body movement is less than the preset body movement threshold, it indicates that the user is in a relatively still sleep state. Therefore, the method for determining the body movement state condition is as follows: the amplitude of the user's body movement is detected based on the body movement characteristics. If the amplitude of body movement is less than the preset body movement threshold (this threshold can be set by taking the upper limit of the 95% confidence interval of the statistical distribution of a large amount of normal sleep body movement data), then the body movement state condition is met.

[0048] The determination of the feature triggering condition varies depending on whether the neck slave device 200 is connected: when the neck slave device 200 is connected, if the laryngeal vibration characteristics meet the preset snoring characteristic frequency band (such as the preset snoring characteristic frequency band is set to 60Hz to 1200Hz), then the feature triggering condition is met. When the neck-worn device 200 is not connected, the rate of decrease in blood oxygen is calculated based on the blood oxygen characteristics, and the sleeping position is determined based on the posture characteristics. If the rate of decrease in blood oxygen is greater than a preset first blood oxygen threshold (e.g., a decrease of more than 2% or 3% per minute) and the posture is a supine sleeping position, then the feature triggering condition is met. In the supine position, gravity causes the base of the tongue to fall back, and the airway cross-sectional area is minimized. If an abnormal rapid decrease in blood oxygen occurs at this time, it is highly likely that the airway has been partially or completely blocked, resulting in insufficient ventilation.

[0049] By combining the above three pre-trigger conditions, the audio acquisition unit 120 will only be awakened when the user is in deep sleep, relatively still, and exhibiting physiological signs of snoring. This maintains the sleep rate of the high-power module at an extremely high level, effectively extending the device's battery life. Simultaneously, the raw audio data collected by the audio acquisition unit 120 after being awakened can be processed and matched locally on the wrist-worn main device 100. After matching, the data is immediately deleted without being uploaded to the cloud, thus eliminating the risk of privacy leaks.

[0050] In the snoring source determination stage, the system uses a multimodal weighted matching algorithm to comprehensively determine the source of snoring.

[0051] Specifically, the system calculates the snoring similarity (e.g., cosine similarity) between audio data and user-pre-acquired snoring templates, the vibration similarity (e.g., cosine similarity) between laryngeal vibration features and user-pre-acquired laryngeal vibration templates, the blood oxygenation similarity (e.g., Pearson correlation coefficient) between blood oxygenation features and pre-acquired snoring blood oxygenation templates, and the laryngeal electromyography similarity (e.g., Pearson correlation coefficient) between laryngeal electromyography features and pre-acquired snoring laryngeal electromyography templates.

[0052] The pre-acquired snoring template and the pre-acquired laryngeal vibration template can be acquired through a specific calibration process (such as actively simulating snoring in a quiet environment) when the user first uses the device, or generated adaptively through high-confidence data from the previous few nights; the pre-acquired snoring blood oxygen template can be a typical curve waveform of blood oxygen saturation changing over time when snoring leads to insufficient ventilation; the pre-acquired snoring laryngeal electromyography template can be a typical electrical activity pattern generated by laryngeal muscles (especially the genioglossus and pharyngeal sphincter) in the process of resisting airway collapse when snoring occurs.

[0053] Subsequently, the system performs a dynamic weighted calculation on the above four similarities (snoring similarity, vibration similarity, blood oxygenation similarity, and laryngeal electromyography similarity) to obtain the individual's snoring similarity:

[0054]

[0055] in, To determine the similarity of snoring to one's own, For audio data, As a snoring template, The vibration characteristics are those of the throat. As a template for throat vibration, Characteristics of blood oxygenation As a template for snoring blood oxygenation, These are the electromyographic characteristics of the laryngeal region. As a laryngeal electromyography template for snoring, These are the weighting coefficients corresponding to audio data, laryngeal vibration characteristics, blood oxygenation characteristics, and laryngeal electromyography characteristics, respectively. .

[0056] In detail, the weights of each feature data are dynamically adjusted based on whether the neck slave device 200 is connected. When the neck slave device 200 is connected, four feature data—audio data, laryngeal vibration characteristics, blood oxygenation characteristics, and laryngeal electromyography characteristics—are used simultaneously for weighting. In a specific parameter configuration example, the weights can be allocated as follows: Even if a partner's snoring results in a high similarity in snoring sounds, the similarity of one's own snoring will still be significantly reduced because there is no vibration in the throat (vibration similarity close to 0) and no electromyographic compensation (laryngeal electromyographic similarity close to 0), thus effectively avoiding misjudgment.

[0057] When the neck slave device 200 is not connected, the matching weight of the laryngeal vibration characteristics and laryngeal electromyographic characteristics is reset to zero. The system then adjusts the weight of the audio data accordingly to achieve adaptive switching between scenarios with and without a neck patch. In a specific parameter configuration example, the weights can be redistributed as follows: At this point, the system relies more on precise matching of acoustic features and auxiliary verification of blood oxygen fluctuations.

[0058] Finally, if the similarity of the user's snoring is greater than the preset similarity threshold, it is determined that the user is snoring; otherwise, it is determined to be the sound of the environment or the sound of the roommate, and the audio acquisition unit 120 is immediately turned off, so that the system returns to a low-power standby state.

[0059] In a practical application scenario of this invention, the preset similarity threshold can be set to 0.75. When the similarity of the user's snoring is greater than or equal to 0.75, it is determined that the user is snoring and the subsequent OSAHS classification and intervention process continues. When the similarity of the user's snoring is less than 0.75, it is determined to be the snoring of the environment / roommates, and the microphone is immediately turned off and returned to a low-power sleep state.

[0060] In the OSAHS classification process, after determining that the user is snoring, the system estimates the AHI (Area of ​​Health) and classifies the condition based on the detected snoring events. The severity of OSAHS is not determined by the decibel level of snoring, but by the frequency of complete airway obstruction (apnea) and partial obstruction (hypopnea) during sleep.

[0061] This process employs a sliding detection window mechanism to analyze sleep data segment by segment throughout the night. Within each preset sliding detection window, it first determines whether an apnea event is met based on the audio data, blood oxygen characteristics, and heart rate characteristics. Then, it determines whether a hypopnea event is met based on the laryngeal vibration characteristics and blood oxygen characteristics. Statistical calculations are performed on apnea and hypopnea events across all sliding detection windows within the preset sleep time range to obtain the apnea-hypopnea index.

[0062] Specifically, within each preset sliding detection window (e.g., 120 seconds), the system determines the snoring interruption time based on audio data and / or laryngeal vibration characteristics, calculates the decrease in blood oxygen based on blood oxygen characteristics, and calculates the increase in heart rate based on heart rate characteristics. If the snoring interruption time within the sliding detection window exceeds a preset interruption threshold (e.g., 10 seconds), the decrease in blood oxygen exceeds a preset second blood oxygen threshold (e.g., 3%), and the increase in heart rate exceeds a preset heart rate threshold (e.g., 10 bpm), it is only considered a genuine apnea event when all three quantitative indicators simultaneously exceed the critical thresholds. For example, if snoring is interrupted for only 12 seconds, but blood oxygen does not decrease and heart rate does not increase, the system will consider this merely a brief breath-hold caused by normal adjustment of the user's breathing rhythm or turning over, rather than pathological apnea.

[0063] Simultaneously, within each sliding detection window, the system determines the decrease in snoring intensity based on laryngeal vibration characteristics, the duration of snoring based on laryngeal vibration characteristics and / or audio data, and the decrease in blood oxygen based on blood oxygen characteristics. If the decrease in snoring intensity within a window exceeds a preset intensity threshold (e.g., 50%, meaning the airflow or vibration amplitude decreases by more than half), the duration of snoring exceeds a preset duration threshold (e.g., 10 seconds), and the decrease in blood oxygen exceeds a preset third blood oxygen threshold (e.g., 3%), then the window is determined to meet the criteria for a hypoventilation event.

[0064] The system counts apnea and hypopnea events across all sliding detection windows within a preset sleep time range, obtaining the total number of events. Then, it calculates the apnea-hypopnea index by ratioing the total number of events to the total duration of the sleep time range.

[0065] in, The Apnea-Hypopnea Index (AHI) is expressed as the average number of apneas and hypopneas occurring during sleep per hour. This represents the total number of apnea events that occurred within the sleep duration. This represents the number of all hypoventilation events occurring within the sleep duration. The total duration of sleep time is expressed in minutes. In this formula, multiplying by 60 is used to convert the denominator to hours so that the unit of AHI is times / hour.

[0066] Based on the calculated AHI estimate, the system categorizes user conditions into four levels—normal, mild, moderate, and severe—according to clinical standards, and outputs corresponding warning information and medical advice, providing clinically valuable OSAHS screening results for consumer-grade devices. Specifically, when AHI < 5, it is considered normal; when 5 ≤ AHI < 15, it is considered mild; when 15 ≤ AHI < 30, it is considered moderate; and when AHI ≥ 30, it is considered severe.

[0067] In the adaptive tiered intervention phase, the system matches the corresponding intervention strategy according to the level of snoring symptoms, prioritizing the use of non-perceptible subthreshold stimulation to avoid waking the user.

[0068] When a hypopnea event is identified, the system generates a first-intensity electrical stimulation command and / or a first-intensity vibration command as intervention instructions. When the snoring event level is mild, the system generates a second-intensity electrical stimulation command and / or a second-intensity vibration command as intervention instructions, where the first intensity value is greater than the second intensity value, meaning the intervention intensity corresponding to the hypopnea event is higher than that for a mild snoring event. When the snoring event level is moderate or severe, the system generates a graded warning command and a linkage device control command.

[0069] At the level of intervention command execution, the wrist vibration unit 140 is used to execute the first intensity value vibration command and the second intensity value vibration command, the subthreshold electrical stimulation unit 230 is used to execute the first intensity value electrical stimulation command and the second intensity value electrical stimulation command, the user terminal display unit 300 is used to execute the graded warning command, and the external ventilator 400 is connected to the wrist main device 100 through wireless communication to execute the linkage device control command.

[0070] The wrist vibration unit 140 outputs physical vibrations of different energy levels by adjusting the driving voltage amplitude and pulse width of the linear motor; the subthreshold electrical stimulation unit 230 applies microcurrents or microvibration energy below the nerve fiber activation threshold that causes pain or significant tactile sensation; the user terminal display unit 300 displays key indicators such as nighttime AHI classification results, longest apnea time, and lowest blood oxygen saturation to the user in the form of intuitive charts, trend lines, and text reports, and provides serious medical advice or lifestyle intervention guidance; after receiving the control command from the wrist main device 100, the external ventilator 400 automatically performs the power-on action and dynamically adjusts the output positive airway pressure (i.e., pressure titration) based on the AHI estimate, blood oxygen drop, and current sleeping position transmitted in real time by the main device.

[0071] For example, when the system detects that the user is in a supine position and their blood oxygen level has dropped sharply, it instructs the ventilator to instantly increase its output pressure to open up the collapsed airway; when the user turns to their side and their breathing becomes stable, it instructs the ventilator to decrease its pressure to improve wearing comfort. This deep cross-device linkage completely breaks the data silo deadlock of traditional home ventilators, which cannot make feedforward adjustments based on the user's real-time deep physiological state, and achieves truly personalized, adaptive, medical-grade closed-loop intervention.

[0072] Through the above-mentioned graded intervention mechanism, the system has achieved a complete intervention gradient, from non-sensory subthreshold stimulation for mild snoring to enhanced intervention for hypoventilation events, and then to cross-device coordinated treatment for moderate to severe OSAHS, effectively solving the problems of traditional interventions that easily wake users and have poor intervention effects.

[0073] In summary, this invention achieves on-demand microphone activation in the physical circuitry through a multi-condition pre-trigger mechanism, significantly extending the device's battery life and eliminating user anxiety about privacy leaks. By dynamically weighting and matching multi-dimensional physiological features, a rigorous snoring verification process is constructed, significantly improving the accuracy of personal snoring recognition in complex environments and overcoming the susceptibility to interference inherent in traditional single acoustic detection. By introducing a sliding detection window and accurately mapping clinical diagnostic criteria, localized high-precision estimation of AHI is achieved. Based on objective symptom grading results, a step-by-step adaptive intervention, from non-invasive subthreshold electrical stimulation to linkage with a home ventilator, is implemented. This effectively blocks airway collapse caused by snoring while greatly reducing damage to normal sleep structure and improving long-term user compliance.

[0074] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0075] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0076] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0077] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A sleep snoring detection and adaptive intervention system based on wrist multimodal data, characterized in that, Including the main wrist device: The wrist-mounted main device includes a multimodal sensing unit, an audio acquisition unit, and a core control unit; The multimodal sensing unit is used to collect the user's multimodal physiological data, and the audio acquisition unit is used to collect audio data after wake-up. The core control unit includes: The pre-trigger wake-up module is used to determine whether multiple pre-trigger conditions are met based on the multimodal physiological data, and wake up the audio acquisition unit when the multiple pre-trigger conditions are met; The snoring source determination module is used to comprehensively determine whether the user is snoring based on the audio data and the multimodal physiological data. The snoring status grading module is used to determine the apnea-hypopnea index based on the audio data and the multimodal physiological data after the user starts snoring, and to determine the level of the user's snoring symptom. The adaptive grading intervention module is used to generate corresponding intervention instructions based on the snoring symptom event level, and call the intervention processing module to execute the intervention instructions.

2. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 1, characterized in that, Also includes neck support devices: The neck slave device includes a miniature piezoelectric vibration sensor and a miniature electromyographic electrode. The neck slave device is fixed to the user's throat on the side of the neck. The wrist master device and the neck slave device are connected to each other through built-in wireless communication units.

3. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 2, characterized in that, The multimodal sensing unit includes a six-axis inertial measurement module, a blood oxygen pulse sensing module, and an electrocardiogram sensing module; The six-axis inertial measurement module is used to collect body motion and posture features, the blood oxygen pulse sensing module is used to collect blood oxygen features, the electrocardiogram sensing module is used to collect heart rate features, the miniature piezoelectric vibration sensor is used to collect laryngeal vibration features, and the miniature electromyography electrode is used to collect laryngeal electromyography features.

4. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 3, characterized in that, When the pre-triggered wake-up module performs the step of determining whether multiple pre-triggered conditions are met based on the multimodal physiological data, it includes: The multiple pre-triggering conditions include sleep state conditions, body movement state conditions, and feature triggering conditions; Sleep staging results are generated based on the body movement characteristics, posture characteristics, and blood oxygenation characteristics. If the sleep staging result is a preset sleep stage, then the sleep state conditions are met. The amplitude of the user's body movement is detected based on the body movement characteristics. If the amplitude of the body movement is less than a preset body movement threshold, the body movement state condition is met. When connecting a neck-mounted device, if the throat vibration characteristics match the preset snoring characteristic frequency band, the characteristic triggering condition is met. When the neck-worn device is not connected, the rate of decrease in blood oxygen is calculated based on the blood oxygen characteristics, and the sleeping posture is determined based on the posture characteristics. If the rate of decrease in blood oxygen is greater than a preset first blood oxygen threshold and the posture characteristics are a supine sleeping posture, then the feature triggering condition is met.

5. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 3, characterized in that, When the snoring source determination module performs the step of comprehensively determining whether the user is snoring based on the audio data and the multimodal physiological data, it includes: Calculate the snoring similarity between the audio data and the user-pre-acquired snoring template; Calculate the vibration similarity between the throat vibration characteristics and the user-pre-acquired throat vibration template; Calculate the blood oxygen similarity between the blood oxygen features and the pre-acquired snoring blood oxygen template; Calculate the laryngeal electromyographic similarity between the laryngeal electromyographic features and the pre-acquired snoring laryngeal electromyographic template; The snoring similarity, vibration similarity, blood oxygenation similarity, and laryngeal electromyography similarity are dynamically weighted and calculated to obtain the individual's snoring similarity. If the similarity of the user's snoring to a preset similarity threshold is greater than the threshold, the user is identified as snoring; otherwise, the sound is identified as environmental or roommate noise, and the audio acquisition unit is turned off.

6. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 3, characterized in that, When the snoring status grading module performs the calculation of the apnea-hypopnea index based on the audio data and the multimodal physiological data, it includes: Within a preset sliding detection window, it is determined whether a sleep apnea event is met based on the audio data, the blood oxygen characteristics, and the heart rate characteristics. In the sliding detection window, it is determined whether a hypoventilation event is met based on the laryngeal vibration characteristics and the blood oxygenation characteristics. The apnea and hypopnea events of all sliding detection windows within a preset sleep time range are statistically calculated to obtain the apnea-hypopnea index.

7. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 6, characterized in that, When the snoring status grading module performs the calculation of the apnea-hypopnea index based on the audio data and the multimodal physiological data, it includes: Within a preset sliding detection window, the snoring interruption time is determined based on the audio data and / or the laryngeal vibration characteristics, the blood oxygen decrease is calculated based on the blood oxygen characteristics, and the heart rate increase is calculated based on the heart rate characteristics. If the snoring interruption time under the sliding detection window is greater than the preset interruption threshold, the blood oxygen drop is greater than the preset second blood oxygen threshold, and the heart rate rise is greater than the preset heart rate threshold, then the sleep apnea event is determined to be satisfied. In the sliding detection window, the decrease in snoring intensity is determined based on the laryngeal vibration characteristics, the duration of snoring is determined based on the laryngeal vibration characteristics and / or audio data, and the decrease in blood oxygen is calculated based on the blood oxygen characteristics. If the decrease in snoring intensity under the sliding detection window is greater than a preset intensity threshold, the duration of snoring is greater than a preset duration threshold, and the decrease in blood oxygen is greater than a preset third blood oxygen threshold, then the hypoventilation event is determined to be met. The total number of events was obtained by counting apnea and hypopnea events in all sliding detection windows within a preset sleep time range. The apnea-hypopnea index is calculated by comparing the total number of events with the total duration of the sleep time range.

8. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 6, characterized in that, The adaptive graded intervention module further includes: when a hypoventilation event is determined to be met, generating a first-intensity electrical stimulation command and / or a first-intensity vibration command as intervention commands.

9. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 8, characterized in that, When the adaptive grading intervention module executes the intervention instruction generated according to the snoring symptom event level, it includes: When the snoring symptom level is mild, a second-intensity electrical stimulation command and / or a second-intensity vibration command are generated as intervention commands. Wherein, the first intensity value in the first intensity value electrical stimulation command is greater than the second intensity value in the second intensity value electrical stimulation command, and the first intensity value in the first intensity value vibration command is greater than the second intensity value in the second intensity value vibration command; When the snoring symptom is classified as moderate or severe, a graded early warning instruction and a linkage device control instruction are generated.

10. The sleep snoring detection and adaptive intervention system based on wrist multimodal data according to claim 9, characterized in that, When the adaptive hierarchical intervention module executes the intervention instruction by calling the intervention processing module, it includes: The intervention module includes a wrist vibration unit located in the main device at the wrist, a subthreshold electrical stimulation unit located in the slave device at the neck, a user terminal display unit, and an external ventilator; The wrist vibration unit is used to execute the first intensity value vibration command and the second intensity value vibration command; The subthreshold electrical stimulation unit is used to execute the first intensity value electrical stimulation command and the second intensity value electrical stimulation command; The user terminal display unit is used to execute the graded early warning command; The external ventilator is wirelessly connected to the wrist-mounted main device and is used to execute control commands for the linked device.

Images