A sleep breathing disorder individualized phenotype identification and closed-loop micro-intervention method and system based on a wearable device

Abstract

The application discloses a sleep breathing disorder individualized phenotype identification and closed-loop micro-intervention method and system based on a wearable device, and belongs to the technical field of sleep breathing disorder management, and comprises the following steps: continuously collecting multi-night data in a natural sleep environment of a user by using a wearable device and its associated devices, identifying suspected breathing disturbance events first, then combining context information such as body position, snoring, blood oxygen drop, heart rate rebound, body movement arousal and sleep stage to determine the sleep breathing disorder phenotype of the user; subsequently, generating an individualized micro-intervention strategy according to the phenotype, and continuously optimizing intervention parameters by using physiological responses after intervention and cross-night control results, and finally outputting associated reports. On the basis of completing risk screening, the application further realizes individualized phenotype identification, closed-loop micro-intervention and cross-night therapeutic effect evaluation, so that the clinical interpretability and landing value of home screening and health management are improved.

Description

Technical Field

[0001] This invention belongs to the technical field of sleep-disordered breathing management, specifically relating to a method and system for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices. Background Technology

[0002] Sleep-disordered breathing, including obstructive sleep apnea, hypopnea, snoring, and sleep-related hypoxia, is common in the population and can lead to daytime sleepiness, decreased attention, and is associated with increased risk of hypertension, cardiovascular disease, and metabolic abnormalities. Currently, polysomnography (PSG) remains the gold standard in clinical practice, but it suffers from problems such as complex equipment, high examination costs, long appointment cycles, and the need for patients to stay overnight in hospitals or designated locations, making it difficult to meet the needs for large-scale, long-term, and continuous home screening and management.

[0003] In recent years, wearable devices have gradually acquired the ability to collect photoplethysmography (PPG), blood oxygen saturation, heart rate, acceleration, body position, skin temperature, and microphone snoring. Some watches, rings, patches, or home sleep monitoring devices can already provide alerts for sleep apnea risk or estimate indicators such as respiratory disturbance index (AHI), hypoxia events, and sleep stages. However, most existing solutions still remain at the screening level of "whether there is a risk" or "estimating AHI / RDI," lacking further attention to questions such as why the user experiences respiratory disturbance, in which body position or sleep stage the disturbance is more severe, and what kind of mild intervention is effective for the user.

[0004] Existing wearable sleep apnea screening programs have the following main shortcomings: (1) Single dimension of results. Existing solutions usually output risk alerts, number of events, AHI estimates or number of times blood oxygen drops, which makes it difficult to further identify different phenotypes such as postural type, snoring obstruction type, hypoxia-dominant type, arousal-dominant type, REM-related type or suspected central type, resulting in a lack of targeted intervention recommendations.

[0005] (2) Poor adaptability of uniform threshold. Different users have significant differences in baseline blood oxygen, resting heart rate, snoring loudness, body movement habits, sleeping posture preferences and sensor wearing quality. If a fixed blood oxygen drop threshold, a fixed snoring threshold or a fixed body movement threshold is used, false alarms or missed alarms are likely to occur.

[0006] (3) Disconnect between detection and intervention. Most existing wearable devices provide risk reports the next day, which cannot provide low-disturbance interventions before or after a user experiences a high-risk respiratory disturbance that night, nor can they assess whether the intervention actually improves the user's respiratory stability.

[0007] (4) Existing postural intervention devices have crude strategies. Some postural therapy products prompt users to leave the supine position through vibration of fixed intensity, but their triggering conditions usually depend only on the position and do not take into account blood oxygenation, heart rate rebound, increased snoring, respiratory micro-movements and sleep status, which can easily lead to ineffective vibration, over-awakening or user tolerance.

[0008] (5) Lack of a closed-loop mechanism for efficacy across nights. Sleep-disordered breathing is characterized by nocturnal fluctuations and multifactorial influences. Single-night testing is insufficient to determine the effectiveness of certain sleep posture adjustments, vibration strategies, pillow height, oral appliances, or CPAP use. There is an urgent need for an efficacy assessment mechanism based on multi-night self-comparison. Summary of the Invention

[0009] The purpose of this invention is to address the aforementioned shortcomings of the prior art by providing a method and system for individualized phenotypic identification and closed-loop micro-intervention of sleep apnea based on wearable devices, in order to solve at least one of the following problems: Existing wearable devices only output sleep apnea risk or AHI estimates, failing to identify the mechanisms and triggering factors of breathing disorders; uniform threshold models are difficult to adapt to different users' baseline blood oxygenation, heart rate response, snoring levels, body movement habits, and wearing differences; detection results are disconnected from real-time nighttime intervention, making it impossible to intervene at low awakening risk when the risk of respiratory disturbance increases; traditional postural therapy or vibration intervention is triggered only by sleeping posture and lacks multimodal physiological feedback, easily leading to ineffective intervention or sleep fragmentation; lack of cross-night self-comparison mechanisms makes it impossible to objectively assess whether different sleeping postures, intervention parameters, or auxiliary treatment devices are effective for specific users.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows: On the one hand, a method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices includes the following steps: S1. Collect multimodal sleep data from individuals; S2. Perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain a comprehensive signal quality weight; S3. Identify an individual's effective sleep period based on multimodal data; S4. Based on sleep stages and body positions, construct individualized nighttime baselines; S5. Calculate individualized deviation characteristics based on individualized nighttime baselines and individual historical baselines; S6. During the effective sleep period, identify candidate respiratory disturbance events based on individualized deviation characteristics; S7. Extract the contextual features of candidate respiratory disturbance events and form an event-level feature vector; S8. Identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events, multimodal data, and temporal correlations between sleep stages. S9. During the effective sleep period, based on the sleep-disorder phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position, generate individualized micro-intervention strategies for different sleep-disorder phenotypes. S10. Adopt a tiered triggering mechanism to implement individualized micro-intervention strategies; S11. Evaluate the response to each individualized micro-intervention strategy and calculate a comprehensive response score; S12. Based on the comprehensive response score, the gradient descent algorithm is used to adaptively update the intervention parameters of the individualized micro-intervention strategy; S13. Compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessment, and generate a report.

[0011] Furthermore, S4 specifically includes: For blood oxygen saturation SpO2, normal heart rate fluctuation range, normal HRV range, common sleep position distribution, quiet snoring level, body movement frequency, respiratory micro-movement amplitude, and respiratory stability under different body positions, the baseline mean and standard deviation were calculated separately for sleep stage and body position.

[0012] Furthermore, S5 specifically includes: Based on the mean and standard deviation of the baseline, the standardized deviation formula is used to compare the real-time collected multimodal sleep data with the historical baseline of the same user, the same sleep stage, and the same body position, and calculate the individualized deviation characteristics.

[0013] Furthermore, S6 specifically includes: Based on individualized deviation characteristics, acceleration body motion signals, and body position signals, a risk prediction model is used to calculate the probability of respiratory disturbance risk by reviewing the temporal features within the window. The probability of respiratory disturbance risk is then smoothed and thresholded, and binary candidate event sequences are extracted. Finally, after duration filtering and merging with adjacent events, candidate respiratory disturbance events within the effective sleep period are identified.

[0014] Furthermore, S7 specifically includes: For each respiratory disturbance candidate event, extract contextual features before, during, and after the event, including body position at the time of the event, snoring trend before the event, lowest blood oxygen saturation SpO2 during the event, heart rate increase after the event, body movement amplitude after the event, event duration, event interval, and sleep stage or nighttime distribution.

[0015] Furthermore, S8 specifically includes: Based on event-level feature vectors, candidate respiratory disturbance events and multimodal data, and the temporal correlation of sleep stages, a confidence calculation model for different sleep-disorder phenotypes is constructed: For position-related types, the event density in supine position and the event density in non-supine position were calculated separately. The phenotypic confidence was calculated by combining the ratio of supine event density to non-supine event density and the difference in hypoxia load with a logistic regression model and a sigmoid function. For the obstructive snoring type, based on the event-level features corresponding to the snoring precursor intensity, heart rate rebound amplitude, and blood oxygen drop amplitude, the phenotypic confidence is calculated using a logistic regression model and a sigmoid function. For REM-related phenotypes, the phenotypic confidence is calculated by combining the event density ratio between REM and non-REM periods with a logistic regression model and a sigmoid function. Based on the confidence level of each phenotype, the phenotype with the highest confidence level is determined as the dominant phenotype, and other phenotypes exceeding the preset threshold are determined as merged phenotypes, thus completing the phenotype identification of sleep-disorder breathing.

[0016] Furthermore, in S9, intervention triggering conditions are constructed based on multi-condition logical combinations: In the formula, The intervention trigger marker at time t; For indicator functions; For sleep mask; Weighting for overall signal quality; This is the signal quality threshold. This represents the probability of respiratory disturbance risk. This is the risk probability threshold; The confidence level of the current phenotype; This is the phenotypic confidence threshold; The trigger time of the last intervention; To intervene in the cooling time; When the intervention triggering conditions are met, the corresponding intervention action is triggered to realize individualized micro-intervention strategies for different phenotypes; Specifically, for individuals with position-related snoring, when a supine position is detected and the risk of respiratory disturbance increases, progressive tactile stimulation is delivered via a watch, ring, chest patch, smart pillow, or mattress to guide the user to slightly turn to their side. For users with snoring obstruction, short bursts of micro-stimulation are delivered before snoring intensifies but before hypoxia occurs. For users with hypoxia-dominant snoring, the frequency of excessive stimulation at night is reduced, and next-day reports and medical advice are prioritized. For users suspected of having central nervous system snoring, frequent positional interventions are avoided and the level of medical assessment prompts is increased.

[0017] Furthermore, S11 specifically includes: Based on each individualized micro-intervention strategy, the intervention response is evaluated from multiple dimensions, and a comprehensive response score is calculated, which is expressed as follows: In the formula, For the first i The overall response score of the intervention; , , , , These are the corresponding weighting coefficients; Mark changes in body position; This refers to the extent of blood oxygen recovery. Reduce the intensity of snoring; For the first i The decrease in the density or number of candidate respiratory disturbance events within the preset observation window after the intervention; For the first i Risk of awakening or sleep fragmentation penalty resulting from the intervention; Subsequently, based on the comparison between the comprehensive response score and the preset effective threshold, the intervention is judged as effective, partially effective, or needs to be downgraded or ineffective.

[0018] Furthermore, S13 specifically includes: An intervention effectiveness score was constructed based on respiratory disturbance index, hypoxia load, snoring load, number of awakenings, and sleep continuity under no-intervention, mild-intervention, intensive-intervention, or different sleep position strategies: In the formula, To score the effectiveness of the intervention; , , , These are the corresponding weighting coefficients; The respiratory disturbance index under no intervention; The respiratory disturbance index under intervention conditions; It is a very small constant; This represents a low oxygen load at baseline. Low oxygen load under intervention conditions; This represents the snoring load at baseline. Snoring load under intervention conditions; The number of awakenings during intervention; The number of awakenings at baseline; The final report includes risk level, main phenotype, phenotype confidence level, high-risk position, high-risk time period, hypoxia load, snoring load, intervention effectiveness, user tolerability, and recommendations for medical attention.

[0019] On the other hand, a personalized phenotypic identification and closed-loop micro-intervention system for sleep-disordered breathing based on wearable devices includes: Signal acquisition module: used to collect multimodal sleep data of an individual; Signal processing module: used to perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain comprehensive signal quality weights; Effective sleep period identification module: used to identify an individual's effective sleep period based on multimodal data; Baseline building module: Used to build individualized nighttime baselines based on sleep stages and body positions; Individualized Deviation Feature Module: Used to calculate individualized deviation features based on individualized nighttime baselines and individual historical baselines; Respiratory disturbance candidate event identification module: used to identify respiratory disturbance candidate events based on individualized deviation characteristics during the effective sleep period; Context feature extraction module: used to extract context features of candidate respiratory disturbance events and form event-level feature vectors; Phenotypic recognition module: used to identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events and multimodal data, and the temporal correlation between sleep stages; Intervention strategy generation module: Used to generate individualized micro-intervention strategies for different sleep-disorder phenotypes based on sleep apnea phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position during effective sleep period; Intervention Execution Module: Used to execute individualized micro-intervention strategies using a tiered triggering mechanism; Intervention Response Assessment Module: Used to assess the response to each individualized micro-intervention strategy and calculate a comprehensive response score; Intervention update module: Used to adaptively update the intervention parameters of individualized micro-intervention strategies based on the comprehensive response score and using the gradient descent algorithm; The efficacy assessment and report generation module is used to compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessments, and generate reports.

[0020] The method and system for individualized phenotypic identification and closed-loop micro-intervention of sleep-disorder based on wearable devices provided by this invention have the following beneficial effects: 1. This invention upgrades from "passive screening" to "proactive management" through multi-night continuous data collection, multimodal respiratory disturbance identification, individualized baseline modeling, phenotypic classification, low-disturbance micro-intervention, adaptive parameter optimization, and cross-night efficacy evaluation.

[0021] 2. This invention upgrades from simple risk screening to individualized phenotype identification; it not only determines whether an individual user has a risk of sleep-disordered breathing, but also further identifies phenotypes such as postural-related, snoring-obstructive, hypoxia-dominant, wakefulness-dominant, REM-related, or suspected central phenotypes, making the report more clinically interpretable and valuable for intervention guidance.

[0022] 3. This invention reduces false alarms and missed alarms by using individualized multi-night baselines; it does not rely entirely on a uniform threshold, but establishes a dynamic baseline based on the user's own baseline blood oxygen, heart rate, snoring, body movement and body position distribution, which can adapt to different ages, body types, wearing habits and device differences.

[0023] 4. This invention achieves a closed loop of detection and intervention; it can perform low-arousal-risk micro-interventions when the risk of respiratory disturbance increases, and automatically adjust the strategy according to the physiological response after intervention, thereby transforming wearable devices from passive monitoring tools into active management tools.

[0024] 5. This invention reduces the risk of awakening during postural therapy; by controlling graded stimulation, cooling time, maximum number of stimulations, and awakening risk, the system avoids sleep fragmentation caused by fixed-intensity vibrations, thereby improving long-term user compliance.

[0025] 6. This invention can perform cross-night efficacy assessment; by comparing multiple nights, it can determine whether different body positions, intervention parameters, auxiliary equipment or treatment methods are effective, providing objective evidence for home health management, remote follow-up and clinical pre- and post-treatment assessment.

[0026] 7. This invention has strong applicability and scalability; it can be implemented individually or in combination in devices such as smartwatches, rings, chest patches, neck patches, smart pillows, mattresses and mobile phones, and is suitable for both consumer-grade health screening and outpatient follow-up and chronic disease management scenarios. Attached Figure Description

[0027] Figure 1 This is a flowchart of the individualized phenotypic identification and closed-loop micro-intervention method for sleep-disordered breathing based on wearable devices in the embodiment.

[0028] Figure 2 This is a schematic diagram illustrating the principle of the individualized phenotypic identification and closed-loop micro-intervention method for sleep-disordered breathing based on wearable devices in this embodiment. Detailed Implementation

[0029] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0030] This embodiment provides a method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices. It utilizes wearable devices and associated devices to continuously collect multi-night data in the user's natural sleep environment. First, it identifies suspected respiratory disturbance events. Then, it combines contextual information such as body position, snoring, decreased blood oxygenation, heart rate rebound, body movement arousal, and sleep stage to determine the user's sleep-disordered breathing phenotype. Subsequently, it generates individualized micro-intervention strategies based on the phenotype and continuously optimizes intervention parameters using post-intervention physiological responses and cross-night control results. Finally, it outputs risk level, main phenotype, effective intervention methods, and medical recommendations. This method can be deployed in smartwatches, smart rings, wristbands, chest patches, neck patches, smart pillows, smart mattresses, mobile phones, or edge computing gateways. It can also be implemented collaboratively by the aforementioned devices and a cloud server. (See reference...) Figure 1 Specifically, it includes the following: S1. Collect multimodal sleep data from individuals; In some embodiments, physiological and motion signals during a user's sleep are collected via at least one wearable device. Physiological signals include one or more of PPG, SpO2, heart rate, HRV, pulse wave amplitude, and skin temperature. Motion signals include one or more of triaxial acceleration, angular velocity, body position, respiratory micro-movements, and turning movements. Simultaneously, one or more of snoring intensity, snoring spectrum, ambient noise, head and neck position, and pillow pressure distribution can be collected via a mobile phone, bedside device, smart pillow, or smart mattress. The multimodal features at second t are then analyzed. Represented as: The multimodal data for the entire night is represented as Where T is the number of seconds recorded throughout the night; where, Let be the blood oxygen saturation at second t; Let t be the heart rate at second t; The heart rate variability at second t; Let be the amplitude of the pulse wave at second t; Let be the triaxial acceleration at second t; Let t be the body position at second t; The snoring characteristics (intensity or spectrum) at second t; The respiratory micro-motion signal at second t; Let t be the skin temperature at second t.

[0031] S2. Perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain a comprehensive signal quality weight; In some embodiments, data from different sampling rates and different device sources are aligned along a unified time axis and aggregated into a feature sequence at 1-second or preset time intervals. Quality indices are calculated separately for PPG, blood oxygen, acceleration, snoring, and body position signals to obtain a comprehensive signal quality weight. ; The overall quality weight is calculated using the following formula: In the formula, , , , , These are the corresponding weighting coefficients; PPG signal quality indicators; This is a quality indicator of blood oxygen saturation signal; For acceleration signal quality indicators; This is an indicator of snoring signal quality. This is a quality indicator for body position signals.

[0032] For the overall signal quality weight Fragments below the threshold are downweighted, interpolated, or removed to avoid misjudgments caused by loose fitting, arm pressure, environmental noise, or short-term sensor distortion.

[0033] S3. Identify an individual's effective sleep period based on multimodal data; In some embodiments, sleep and wakefulness periods are determined based on acceleration, heart rate, HRV, skin temperature, PPG stability, and body movement frequency, and further, light sleep, deep sleep, and REM sleep tendency stages can be estimated. The sleep mask is denoted as... When time t is within the effective sleep period When time t is in the awakening period, preparation period, or low-confidence segment... ; Effective total sleep time Calculate using the following formula: S4. Based on sleep stages and body positions, construct individualized nighttime baselines; In some embodiments, an individualized baseline is established based on data from at least one night or several consecutive nights of user data, including calculating the mean and standard deviation of the baseline for blood oxygen saturation SpO2, normal heart rate fluctuation range, normal HRV range, common sleep position distribution, quiet snoring level, body movement frequency, respiratory micro-movement amplitude, and respiratory stability under different body positions, respectively, according to sleep stage and body position. For any feature f Calculate the baseline mean and standard deviation according to sleep stage s and body position p respectively: Where u represents the user, This represents the historical nighttime data set used for modeling. , Let be the effective weight of the sample at time t on night n under a specified sleep stage s and body position p. Let t be the sleep mask for the nth night at time t. It is set to 1 when the person is in an effective sleep period, and 0 otherwise. This refers to the sleep stage label at time t on night n. The target sleep stage currently used for baseline statistics. This represents the body position label at time t on night n. The target body position currently used for baseline statistics; Let f be the observed value of the characteristic f at time t on the nth night; Indicates user Characteristics under sleep stage s and body position p f The baseline mean; Indicates user Characteristics under sleep stage s and body position p f The baseline standard deviation; This is an indicator function.

[0034] S5. Calculate individualized deviation characteristics based on individualized nighttime baselines and individual historical baselines; In some embodiments, instead of relying directly on a fixed threshold, the standardized deviation formula is used based on the mean and standard deviation of the baseline to compare the real-time collected multimodal sleep data with the historical baseline of the same user, the same sleep stage, and the same body position, and to calculate the individualized deviation characteristics. Taking blood oxygen, heart rate, snoring, and respiratory micro-movements as examples: in, To prevent extremely small constants with a denominator of zero; This refers to the deviation characteristics of blood oxygen saturation. The historical baseline mean of blood oxygen saturation for user u during sleep stage s and body position p. The historical baseline standard deviation of blood oxygen saturation for user u during sleep stage s and body position p; This is a characteristic of heart rate deviation. The historical baseline mean of user u's heart rate during sleep stage s and body position p. The baseline standard deviation of user u's heart rate during sleep stage s and body position p; This refers to the deviation characteristics of snoring signals. The historical baseline mean of snoring level for user u during sleep stage s and body position p. The historical baseline standard deviation of snoring level for user u during sleep stage s and body position p; This is a deviation characteristic of respiratory micromovements. The historical baseline mean of respiratory micromovements of user u during sleep stage s and body position p. This represents the baseline standard deviation of the respiratory micromovement history of user u during sleep stage s and body position p. This embodiment can accommodate differences in baseline blood oxygenation, snoring levels, and body movement habits among different users.

[0035] S6. During the effective sleep period, identify candidate respiratory disturbance events based on individualized deviation characteristics; In some embodiments, based on individualized deviation characteristics, acceleration body motion signals, and body position signals, a risk prediction model is used to calculate the probability of respiratory disturbance risk by reviewing the temporal features within the window; then, the probability of respiratory disturbance risk is smoothed and thresholded, and a binary candidate event sequence is extracted; finally, after duration filtering and merging with adjacent events, candidate respiratory disturbance events within the effective sleep period are identified. The risk prediction model in this embodiment can be a rule-based model, a machine learning model, a deep learning model, or a hybrid model, wherein the respiratory disturbance risk probability Represented as: in, For the Sigmoid function, For the risk prediction model, L is the length of the lookback window; To review the deviation characteristics of blood oxygen saturation within the window. To review the deviation characteristics of heart rate within the window, To review the deviation characteristics of the snoring signal within the window, To review the deviation characteristics of the respiratory micro-movements within the window. To review the deviation characteristics of the three-axis acceleration within the window, To review the deviation characteristics of the body position within the window; probability of disturbance risk Smoothing and thresholding are performed to obtain a binary candidate event sequence. : In the formula, To achieve a smooth window width of half, the total length of the smooth window is 2K+1. Let be the probability of respiratory disturbance risk at time t+i. The binary candidate event label at time t, This is the threshold for the probability of respiratory disturbance risk. For continuous A set of candidate events consisting of fragments = 1. Let j be the start time of the j-th candidate event. Let j be the end time of the j-th candidate event; Segments with a duration shorter than the preset minimum duration are filtered out, and segments with an adjacent interval shorter than the preset interval are merged.

[0036] S7. Extract the contextual features of candidate respiratory disturbance events and form an event-level feature vector; In some embodiments, for each respiratory disturbance candidate event Extract contextual features before, during, and after the event, including body position at the time of the event, snoring trend before the event, lowest blood oxygen saturation SpO2 during the event, increase in heart rate after the event, increase in body movement after the event, duration of the event, event interval, and sleep stage or nighttime time distribution.

[0037] The duration of the event, the magnitude of the drop in blood oxygen, the magnitude of the heart rate rebound, and the intensity of the snoring precursor can be calculated as follows: In the formula, Let be the duration of the j-th candidate event. Let represent the decrease in blood oxygen saturation for the j-th candidate event relative to the pre-event baseline. The pre-event blood oxygen level value corresponding to the j-th candidate event. Let $\frac{j}{j}$ be the heart rate bounce amplitude after the $j$-th candidate event. The length of the time window used to calculate the mean before and after the event. The average snoring precursor intensity is defined as H seconds prior to the occurrence of the j-th candidate event.

[0038] S8. Identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events, multimodal data, and temporal correlations between sleep stages. In some embodiments, confidence calculation models for different sleep-disorder phenotypes are constructed based on event-level feature vectors, candidate respiratory disturbance events and multimodal data, and the temporal correlation of sleep stages. The phenotypes include postural-related type, snoring obstruction type, hypoxia-dominant type, wakefulness-dominant type, REM-related type, and suspected central type. For position-related phenotypes, the event density in the supine and non-supine positions was calculated separately. The phenotypic confidence was then calculated using the ratio of supine to non-supine event density, the difference in hypoxia load, and a logistic regression model combined with the sigmoid function. This confidence level is expressed as follows: In the formula, This represents the density of respiratory events in the supine position. This represents the number of respiratory disturbance events detected in the supine position. The effective sleep duration in the supine position is calculated in seconds. Density of respiratory events in non-supine position. This represents the number of respiratory disturbance events detected in non-supine positions. The effective sleep duration in non-supine position is calculated in seconds. For position-related confidence levels; , , These are the corresponding weighting coefficients; Supine position with low oxygen load. Low oxygen load in non-supine position; For obstructive snoring, based on the event-level features corresponding to snoring precursor intensity, heart rate rebound amplitude, and blood oxygen saturation drop amplitude, the phenotypic confidence is calculated using a logistic regression model and the sigmoid function, and is expressed as follows: In the formula, For snoring obstruction type confidence level; , , , These are the corresponding weighting coefficients; The event-level characteristic mean of the snoring precursor intensity. The mean of the event-level features of the heart rate rebound amplitude. This represents the event-level characteristic of the magnitude of the decrease in blood oxygen saturation; For REM-related phenotypes, the phenotypic confidence is calculated by combining the event density ratio between REM and non-REM periods with a logistic regression model and the sigmoid function, and is expressed as follows: In the formula, For REM correlation confidence levels; , For the corresponding weighting coefficients, The density of respiratory events during REM sleep. Density of respiratory events during non-REM phase; Subsequently, based on the confidence level of each phenotype, the phenotype with the highest confidence level is determined as the dominant phenotype, and other phenotypes exceeding the preset threshold are determined as merged phenotypes, thus completing the phenotype identification of sleep-disorder breathing.

[0039] S9. During the effective sleep period, based on the sleep-disorder phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position, generate individualized micro-intervention strategies for different sleep-disorder phenotypes. In some embodiments, intervention triggering conditions are first constructed based on a combination of multiple conditional logics: In the formula, The intervention trigger marker at time t; This is the signal quality threshold. This is the risk probability threshold; The confidence level of the current phenotype; This is the phenotypic confidence threshold; The trigger time of the last intervention; To intervene in the cooling time; When the intervention triggering conditions are met, the corresponding intervention action is triggered to realize individualized micro-intervention strategies for different phenotypes; Specifically, for individuals with position-related snoring, when a supine position is detected and the risk of respiratory disturbance increases, progressive tactile stimulation is delivered via a watch, ring, chest patch, smart pillow, or mattress to guide the user to slightly turn to their side. For users with snoring obstruction, short bursts of micro-stimulation are delivered before snoring intensifies but before hypoxia occurs. For users with hypoxia-dominant snoring, the frequency of excessive stimulation at night is reduced, and next-day reports and medical advice are prioritized. For users suspected of having central nervous system snoring, frequent positional interventions are avoided and the level of medical assessment prompts is increased.

[0040] S10. Adopt a tiered triggering mechanism to implement individualized micro-intervention strategies; In some embodiments, the micro-intervention employs a graded triggering mechanism, initially delivering low-intensity, short-duration stimulation. If no postural change or physiological recovery occurs, the stimulation intensity, duration, or number of repetitions is gradually increased. The intensity of the k-th stimulation can be increased according to the following formula: In the formula, The stimulation intensity of the kth micro-intervention is... To the maximum allowed stimulus intensity, The initial stimulus intensity, Let be the increment step size for each iteration. The duration of the stimulation during the k-th micro-intervention. To the maximum allowed duration of the output, The duration of the initial stimulus. This represents the incrementing duration step.

[0041] The system is configured with a cooldown time, maximum number of nighttime stimulations, arousal risk threshold, and user tolerance parameters. Intervention is paused when significant arousal, continuous body movement, or abnormally elevated heart rate is detected.

[0042] Awakening Risk It can be represented as: In the formula, , , , For the corresponding weighting coefficients; when When the threshold is exceeded, the system stops or reduces stimulation.

[0043] S11. Evaluate the response to each individualized micro-intervention strategy and calculate a comprehensive response score; In some embodiments, response indicators are calculated after each microintervention, including whether body position changes, whether snoring decreases, whether SpO2 recovers, whether heart rate rebound weakens, whether candidate event density decreases in the following 5 to 10 minutes, and whether awakening or sleep fragmentation occurs. For the first The comprehensive response score for this intervention can be expressed as: In the formula, For the first i The overall response score of the intervention; , , , , These are the corresponding weighting coefficients; Mark changes in body position; This refers to the extent of blood oxygen recovery. Reduce the intensity of snoring; The decrease in the density or number of candidate respiratory disturbance events within the preset observation window after the i-th intervention; The penalty for awakening risk or sleep fragmentation caused by the i-th intervention; Subsequently, based on the comparison between the comprehensive response score and the preset effective threshold, if If the intervention is above the effective threshold, it is considered effective; if the number of events decreases but the risk of arousal is high, it is considered partially effective or needs to be downgraded; if there is no change in body position, no physiological recovery, and no decrease in events, it is considered ineffective.

[0044] S12. Based on the comprehensive response score, the gradient descent algorithm is used to adaptively update the intervention parameters of the individualized micro-intervention strategy; In some embodiments, the trigger threshold, vibration intensity, vibration rhythm, duration, cooling interval, and applicable body position are updated based on the comprehensive response score. When low-intensity stimulation can stably guide a change in body position without causing arousal, the stimulation intensity is maintained or reduced; when multiple stimuli are ineffective, the trigger condition is increased or such intervention is stopped; when stimulation causes arousal, the intensity is reduced or the cooling interval is extended. Parameter updates may take the following forms: In the formula, This is the set of individualized micro-intervention parameters after the (n+1)th iteration. This is the set of individualized micro-intervention parameters for the nth iteration, which includes one or more of the following: trigger threshold, vibration intensity, vibration rhythm, stimulation duration, cooling interval, and applicable body position. For the objective function Relative to parameters The gradient operator; Indicates the learning rate. Indicates the target response score. This represents the objective function that balances event reduction and sleep continuity.

[0045] S13. Compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessment, and generate a report; In some embodiments, data from different nights or different time periods of the same night are compared to compare respiratory disturbance index, hypoxia load, snoring load, number of awakenings and sleep continuity under no intervention, light intervention, intensive intervention or different sleep position strategies. Respiratory Disturbance Index and low oxygen load It can be calculated using the following formula: The intervention effectiveness score is expressed as follows: In the formula, This represents the total number of respiratory disturbance events. This is the baseline value for blood oxygen saturation; To score the effectiveness of the intervention; , , , These are the corresponding weighting coefficients; The respiratory disturbance index under no intervention; The respiratory disturbance index under intervention conditions; It is a very small constant; This represents a low oxygen load at baseline. Low oxygen load under intervention conditions; This represents the snoring load at baseline. Snoring load under intervention conditions; The number of awakenings during intervention; The number of awakenings at baseline; The final report includes risk level, main phenotype, phenotype confidence level, high-risk position, high-risk time period, hypoxia load, snoring load, intervention effectiveness, user tolerability, and recommendations for medical attention.

[0046] This embodiment also provides a personalized phenotypic identification and closed-loop micro-intervention system for sleep apnea based on wearable devices, referencing... Figure 2 ,include: Signal acquisition module: used to collect multimodal sleep data of an individual; Signal processing module: used to perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain comprehensive signal quality weights; Effective sleep period identification module: used to identify an individual's effective sleep period based on multimodal data; Baseline building module: Used to build individualized nighttime baselines based on sleep stages and body positions; Individualized Deviation Feature Module: Used to calculate individualized deviation features based on individualized nighttime baselines and individual historical baselines; Respiratory disturbance candidate event identification module: used to identify respiratory disturbance candidate events based on individualized deviation characteristics during the effective sleep period; Context feature extraction module: used to extract context features of candidate respiratory disturbance events and form event-level feature vectors; Phenotypic recognition module: used to identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events and multimodal data, and the temporal correlation between sleep stages; Intervention strategy generation module: Used to generate individualized micro-intervention strategies for different sleep-disorder phenotypes based on sleep apnea phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position during effective sleep period; Intervention Execution Module: Used to execute individualized micro-intervention strategies using a tiered triggering mechanism; Intervention Response Assessment Module: Used to assess the response to each individualized micro-intervention strategy and calculate a comprehensive response score; Intervention update module: Used to adaptively update the intervention parameters of individualized micro-intervention strategies based on the comprehensive response score and using the gradient descent algorithm; The efficacy assessment and report generation module is used to compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessments, and generate reports.

[0047] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of this patent.

Claims

1. A method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices, characterized in that, Includes the following steps: S1. Collect multimodal sleep data from individuals; S2. Perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain a comprehensive signal quality weight; S3. Identify an individual's effective sleep period based on multimodal data; S4. Based on sleep stages and body positions, construct individualized nighttime baselines; S5. Calculate individualized deviation characteristics based on individualized nighttime baselines and individual historical baselines; S6. During the effective sleep period, identify candidate respiratory disturbance events based on individualized deviation characteristics; S7. Extract the contextual features of candidate respiratory disturbance events and form an event-level feature vector; S8. Identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events, multimodal data, and temporal correlations between sleep stages. S9. During the effective sleep period, based on the sleep-disorder phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position, generate individualized micro-intervention strategies for different sleep-disorder phenotypes. S10. Adopt a tiered triggering mechanism to implement individualized micro-intervention strategies; S11. Evaluate the response to each individualized micro-intervention strategy and calculate a comprehensive response score; S12. Based on the comprehensive response score, the gradient descent algorithm is used to adaptively update the intervention parameters of the individualized micro-intervention strategy; S13. Compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessment, and generate a report.

2. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, S4 specifically includes: For blood oxygen saturation SpO2, normal heart rate fluctuation range, normal HRV range, common sleep position distribution, quiet snoring level, body movement frequency, respiratory micro-movement amplitude, and respiratory stability under different body positions, the baseline mean and standard deviation were calculated separately for sleep stage and body position.

3. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 2, characterized in that, S5 specifically includes: Based on the mean and standard deviation of the baseline, the standardized deviation formula is used to compare the real-time collected multimodal sleep data with the historical baseline of the same user, the same sleep stage, and the same body position, and calculate the individualized deviation characteristics.

4. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, S6 specifically includes: Based on individualized deviation characteristics, acceleration body motion signals, and body position signals, a risk prediction model is used to calculate the probability of respiratory disturbance risk by reviewing the temporal features within the window. The probability of respiratory disturbance risk is then smoothed and thresholded, and binary candidate event sequences are extracted. Finally, after duration filtering and merging with adjacent events, candidate respiratory disturbance events within the effective sleep period are identified.

5. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, Specifically, S7 includes: For each respiratory disturbance candidate event, extract contextual features before, during, and after the event, including body position at the time of the event, snoring trend before the event, lowest blood oxygen saturation SpO2 during the event, heart rate increase after the event, body movement amplitude after the event, event duration, event interval, and sleep stage or nighttime distribution.

6. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, S8 specifically includes: Based on event-level feature vectors, candidate respiratory disturbance events and multimodal data, and the temporal correlation of sleep stages, a confidence calculation model for different sleep-disorder phenotypes is constructed: For position-related types, the event density in supine position and the event density in non-supine position were calculated separately. The phenotypic confidence was calculated by combining the ratio of supine event density to non-supine event density and the difference in hypoxia load with a logistic regression model and a sigmoid function. For the obstructive snoring type, based on the event-level features corresponding to the snoring precursor intensity, heart rate rebound amplitude, and blood oxygen drop amplitude, the phenotypic confidence is calculated using a logistic regression model and a sigmoid function. For REM-related phenotypes, the phenotypic confidence is calculated by combining the event density ratio between REM and non-REM periods with a logistic regression model and a sigmoid function. Based on the confidence level of each phenotype, the phenotype with the highest confidence level is determined as the dominant phenotype, and other phenotypes exceeding the preset threshold are determined as merged phenotypes, thus completing the phenotype identification of sleep-disorder breathing.

7. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, In S9, intervention triggering conditions are constructed based on a combination of multiple conditional logics: In the formula, The intervention trigger marker at time t; For indicator functions; For sleep mask; Weighting for overall signal quality; This is the signal quality threshold. This represents the probability of respiratory disturbance risk. This is the risk probability threshold; The confidence level of the current phenotype; This is the phenotypic confidence threshold; The trigger time of the last intervention; To intervene in the cooling time; When the intervention triggering conditions are met, the corresponding intervention action is triggered to realize individualized micro-intervention strategies for different phenotypes; Specifically, for individuals with position-related snoring, when a supine position is detected and the risk of respiratory disturbance increases, progressive tactile stimulation is delivered via a watch, ring, chest patch, smart pillow, or mattress to guide the user to slightly turn to their side. For users with snoring obstruction, short bursts of micro-stimulation are delivered before snoring intensifies but before hypoxia occurs. For users with hypoxia-dominant snoring, the frequency of excessive stimulation at night is reduced, and next-day reports and medical advice are prioritized. For users suspected of having central nervous system snoring, frequent positional interventions are avoided and the level of medical assessment prompts is increased.

8. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep apnea based on wearable devices according to claim 7, characterized in that, S11 specifically includes: Based on each individualized micro-intervention strategy, the intervention response is evaluated from multiple dimensions, and a comprehensive response score is calculated, which is expressed as follows: In the formula, For the first i The overall response score of the intervention; , , , , These are the corresponding weighting coefficients; Mark changes in body position; This refers to the extent of blood oxygen recovery. Reduce the intensity of snoring; For the first i The decrease in the density or number of candidate respiratory disturbance events within the preset observation window after the intervention; For the first i Risk of awakening or sleep fragmentation penalty resulting from the intervention; Subsequently, based on the comparison between the comprehensive response score and the preset effective threshold, the intervention is judged as effective, partially effective, or needs to be downgraded or ineffective.

9. The method for individualized phenotypic identification and closed-loop micro-intervention of sleep-disordered breathing based on wearable devices according to claim 1, characterized in that, S13 specifically includes: An intervention effectiveness score was constructed based on respiratory disturbance index, hypoxia load, snoring load, number of awakenings, and sleep continuity under no-intervention, mild-intervention, intensive-intervention, or different sleep position strategies: In the formula, To score the effectiveness of the intervention; , , , These are the corresponding weighting coefficients; The respiratory disturbance index under no intervention; The respiratory disturbance index under intervention conditions; It is a very small constant; This represents a low oxygen load at baseline. Low oxygen load under intervention conditions; This represents the snoring load at baseline. Snoring load under intervention conditions; The number of awakenings during intervention; The number of awakenings at baseline; The final report includes risk level, main phenotype, phenotype confidence level, high-risk position, high-risk time period, hypoxia load, snoring load, intervention effectiveness, user tolerability, and recommendations for medical attention.

10. A personalized phenotypic identification and closed-loop micro-intervention system for sleep apnea based on wearable devices, characterized in that, include: Signal acquisition module: used to collect multimodal sleep data of an individual; Signal processing module: used to perform time synchronization processing and signal quality assessment on multimodal sleep data to obtain comprehensive signal quality weights; Effective sleep period identification module: used to identify an individual's effective sleep period based on multimodal data; Baseline building module: Used to build individualized nighttime baselines based on sleep stages and body positions; Individualized Deviation Feature Module: Used to calculate individualized deviation features based on individualized nighttime baselines and individual historical baselines; Respiratory disturbance candidate event identification module: used to identify respiratory disturbance candidate events based on individualized deviation characteristics during the effective sleep period; Context feature extraction module: used to extract context features of candidate respiratory disturbance events and form event-level feature vectors; Phenotypic recognition module: used to identify sleep-disordered breathing phenotypes based on event-level feature vectors, respiratory disturbance candidate events and multimodal data, and the temporal correlation between sleep stages; Intervention strategy generation module: Used to generate individualized micro-intervention strategies for different sleep-disorder phenotypes based on sleep apnea phenotype, comprehensive signal quality weight, respiratory disturbance risk probability, and current body position during effective sleep period; Intervention Execution Module: Used to execute individualized micro-intervention strategies using a tiered triggering mechanism; Intervention Response Assessment Module: Used to assess the response to each individualized micro-intervention strategy and calculate a comprehensive response score; Intervention update module: Used to adaptively update the intervention parameters of individualized micro-intervention strategies based on the comprehensive response score and using the gradient descent algorithm; The efficacy assessment and report generation module is used to compare sleep-related breathing indicators under different intervention strategies or different sleeping position strategies, conduct cross-night self-comparison efficacy assessments, and generate reports.

Images