Sleep apnea monitoring system based on contactless intervention

Abstract

The application discloses a sleep apnea monitoring system based on non-contact intervention and belongs to the technical field of medical health monitoring.The application solves the problems of the PSG method in the prior art, such as complicated operation, high cost, and the discomfort brought to users by household contact devices, and influences natural sleep, physiological signals are collected by a non-contact biological radar sensor, an optimal threshold is calculated by comprehensively considering an initial threshold, an auxiliary threshold and environmental factors, and the accuracy of apnea and hypopnea event detection is improved, in combination with an apnea hypopnea index, an event duration factor, a blood oxygen drop factor and an event type factor, a severity score is calculated, patients are divided into mild, moderate and severe, according to the severity, grade and snoring sound category of real-time events, an optimal intervention measure is selected from an intervention strategy library, individualized intervention is realized, the interference on patient sleep is reduced, and the sleep quality and life quality are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical health monitoring, in particular to a sleep apnea monitoring system based on non-contact intervention. BACKGROUND

[0002] Sleep apnea syndrome (SAS) is a common sleep disorder characterized by repeated respiratory interruptions during sleep. Long-term suffering from this disease can lead to high blood pressure, heart disease, stroke, daytime sleepiness and other serious consequences, seriously affecting the quality of life and safety of patients.

[0003] Currently, the gold standard for diagnosing the disease is polysomnography (PSG) monitoring, which requires a large number of electrodes and sensors to be attached to the patient's body. The process is cumbersome, the environment is unfamiliar, the cost is high, and the patient's sleep may be disturbed due to the restraint of the lead, affecting the accuracy of the monitoring results, and it is not suitable for long-term, home-based routine screening and monitoring. In addition, existing home monitoring devices are mostly contact type, such as wrist devices, finger clip oximeters, etc., which still cause discomfort to users and affect natural sleep.

[0004] Therefore, the existing needs are not met, and for this we propose a sleep apnea monitoring system based on non-contact intervention. SUMMARY

[0005] The purpose of the present application is to provide a sleep apnea monitoring system based on non-contact intervention, which calculates the threshold confidence of each patient by synthesizing the initial threshold value, auxiliary threshold value and environmental factors through an adaptive threshold function, selects the initial threshold value with the highest confidence as the optimal threshold value, improves the accuracy and reliability of apnea and hypopnea event detection, and provides accurate monitoring and diagnosis for patients. Combined with the apnea hypopnea index, event duration factor, blood oxygen drop factor and event type factor, the severity score is calculated and graded. According to the severity, grade and snoring category of real-time events, the optimal intervention measure is selected from the intervention strategy library to achieve personalized intervention, reduce the disturbance to the patient's sleep, improve the sleep quality and quality of life, and solve the problems raised in the above background technology.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] The sleep apnea monitoring system based on non-contact intervention comprises:

[0008] A non-contact signal acquisition unit configured to acquire biological signals including respiratory waveform, heart beat waveform and body movement information based on a biological radar sensor, and to monitor snoring data and sleep environment data in real time based on a microphone array and a temperature and humidity sensor;

[0009] The sleep data analysis unit is configured to collect historical records, extract biological feature parameters related to apnea from historical biological signals, including but not limited to respiratory amplitude, respiratory interval, respiratory rhythm disorder degree, heart rate variability and blood oxygen saturation; extract apnea and hypopnea events from the biological feature parameters based on a preset threshold as training data;

[0010] The apnea recognition unit is configured to build a convolutional neural network model, learn the biological feature parameters of the apnea and hypopnea events in the training data using the convolutional neural network model, and identify the apnea and hypopnea events therefrom;

[0011] The apnea intervention unit is configured to calculate the number of apnea and hypopnea events occurring per unit time, and evaluate the severity accordingly; when an apnea event is determined to have occurred, the corresponding intervention measures are automatically triggered in combination with the severity.

[0012] Further, the sleep data analysis unit comprises:

[0013] The threshold setting module is configured to select biological signals of multiple patients from historical biological signals, extract biological feature parameters related to apnea, and calculate the median of the biological feature parameters of each patient as an initial threshold;

[0014] Obtain biological signals and snoring data of multiple patients without significant abnormalities within a specified unit of time as auxiliary feature parameters;

[0015] Calculate the baseline level of the auxiliary feature parameters of each patient as an auxiliary threshold;

[0016] Obtain sleep environment data corresponding to the biological signals of the multiple patients as environmental factors;

[0017] Introduce a weighted adaptive threshold function to calculate the initial threshold, auxiliary threshold and environmental factors of the multiple patients to obtain the threshold confidence of the multiple patients;

[0018] Sort the threshold confidence of the multiple patients, extract the patient with the highest threshold confidence, and use the initial threshold of the patient as the optimal threshold for subsequent detection of apnea and hypopnea events.

[0019] Further, introduce a weighted adaptive threshold function to calculate the initial threshold, auxiliary threshold and environmental factors of the multiple patients to obtain the threshold confidence of the multiple patients, comprising:

[0020] Based on the initial threshold, auxiliary threshold and environmental factors, the initial weight is obtained based on the historical influence degree of apnea;

[0021] Optimize the initial weight using the weighted adaptive threshold function to obtain an optimized weight;

[0022] The initial threshold value, the auxiliary threshold value and the environmental factor are integrated with the corresponding optimization weight to obtain a threshold confidence of each patient.

[0023] Further, the sleep data analysis unit further comprises:

[0024] A fluctuation range of the threshold confidence of each patient in a state without sleep is obtained;

[0025] The threshold confidence of the patient whose fluctuation range is greater than a preset fluctuation threshold value is removed to obtain a first threshold confidence set;

[0026] Physiological feature data of each patient in the first threshold confidence set is obtained, and the physiological feature data includes patient disease data;

[0027] Comorbid disease data related to sleep apnea is extracted from a preset disease database;

[0028] The disease data of each patient is compared with the comorbid disease data, and the patient corresponding to the disease data matching the comorbid disease data is taken as a corrected patient;

[0029] The threshold confidence of the corrected patient is removed from the first threshold confidence set to obtain a second threshold confidence set;

[0030] The threshold confidences in the second threshold confidence set are sorted, the patient with the highest threshold confidence is extracted, and the initial threshold value corresponding to the patient is taken as a first threshold value;

[0031] Physiological feature data and behavior habit data of the patient corresponding to the first threshold value are obtained, the physiological feature data includes patient body type of the patient, and the behavior habit data includes patient sleep posture of the patient;

[0032] A first influence threshold value of the patient is determined based on a deviation degree of the patient body type of the patient from a median of a standard body type range interval;

[0033] A second influence threshold value of the patient is determined based on a deviation degree of the patient sleep posture of the patient from a standard sleep posture;

[0034] The first threshold value of the patient is threshold-optimized by using the first influence threshold value and the second influence threshold value to obtain a second threshold value;

[0035] The sleep stage of the patient is determined based on the physiological feature data of the patient, and the second threshold value is optimized in different stages based on different sleep stages to obtain an optimized optimal threshold value, the optimal threshold value including an optimized threshold value of each sleep stage.

[0036] Further, the sleep apnea intervention unit comprises:

[0037] The evaluation criterion setting module is configured to calculate the number of apnea and hypopnea events of multiple patients in a unit of time in historical biological signals as an apnea hypopnea index;

[0038] The duration of each apnea and hypopnea event of each patient is recorded, and the average duration of all events is calculated as an event duration factor;

[0039] The oxygen saturation decrease trend of each patient is calculated by the biological radar sensor, and the average oxygen decrease amplitude of all events is calculated as an oxygen decrease factor;

[0040] The snoring sound data of multiple patients is analyzed to distinguish the categories of the snoring sound data, including obstructive and central, as an event type factor; and the obstructive is given a higher weight and the central is given a lower weight;

[0041] The weighted summation method is used to calculate the apnea hypopnea index respectively with the event duration factor, the oxygen decrease factor and the event type factor to obtain a severity score;

[0042] According to the severity scores of multiple patients, they are divided into mild, moderate and severe.

[0043] Further, the apnea intervention unit further comprises:

[0044] The intervention module is configured to, when determining that an apnea event occurs, obtain the severity of the real-time event, the severity level and the snoring sound category to which the event belongs, and combine them to select the optimal intervention measure from the intervention strategy library;

[0045] The feedback module is configured to, after the intervention is implemented, continue to monitor the user's respiratory signal through the biological radar sensor to form a closed-loop feedback; if the respiration has not returned to normal within a set time, an escalation strategy is started; if it has returned to normal, the intervention is slowly withdrawn after a period of time, so that the patient benefits without feeling, and sleep interruption is minimized.

[0046] Further, the sleep data analysis unit further comprises:

[0047] The data processing module is configured to pre-process the biological signals, including but not limited to filtering, noise reduction and separation, to obtain pre-processed biological signals;

[0048] The data division module is configured to filter out apnea and hypopnea events from biological feature parameters based on a preset threshold as a training set of the convolutional neural network model for training the convolutional neural network model; and the pre-processed biological signals are used as a test set for verifying whether the prediction result of the convolutional neural network model is accurate.

[0049] Further, the intervention measures include, but are not limited to, inducing the subject to change the sleeping posture through a vibration device, stimulating the laryngeal muscle to expand the airway through an electric stimulation device, adjusting the hardness and shape of the mattress through an intelligent mattress to improve the respiratory condition, adjusting the indoor temperature, humidity and ventilation through an environmental control device to optimize the sleeping environment.

[0050] Further, it further comprises:

[0051] A report generation unit configured to automatically generate a visual sleep health report based on the sleep data analysis results of the patient per night;

[0052] A user interaction unit configured to provide an interactive interface for the patient's family members and medical personnel through a mobile terminal application, to display the current respiratory waveform, heart beat waveform, body movement information, snoring data and sleep environment data, and historical sleep health report in real time, and to realize the visualization of monitoring data and intervention feedback.

[0053] Further, it further comprises:

[0054] An alarm unit configured to generate and push a hierarchical alarm information to the preset guardian mobile phone application or medical platform when a high-risk event is detected, while triggering the intervention measures;

[0055] A recording unit configured to record all alarm events and their intervention measure processing states after the intervention measures are implemented;

[0056] A storage unit configured to store the biological signal data and corresponding event labels of the patient after long-term intervention treatment, to form a data set;

[0057] An optimization unit configured to use the data set to perform incremental learning and fine-tuning on the threshold of the biological feature parameters, the severity score and the convolutional neural network model on a regular basis, to adapt to the physiological changes of the patient;

[0058] A collaborative management unit configured to support the patient's family members to adjust the sensitivity and notification mode of the intervention measures and to confirm, correct or label the events recorded by the system based on personal preferences through a mobile terminal application.

[0059] Compared with the prior art, the present application has the following advantages:

[0060] 1. In this invention, an adaptive threshold function is used to comprehensively consider the initial threshold, auxiliary threshold, and environmental factors to calculate the threshold confidence level for each patient. The initial threshold of the patient with the highest threshold confidence level is then selected as the optimal threshold. This improves the clinical applicability and reliability of the threshold, thereby ensuring the effectiveness of subsequent identification of biometric parameters and enhancing the accuracy and reliability of apnea and hypoventilation event detection. This provides patients with more precise monitoring and diagnosis, helps to promptly detect and intervene in sleep-disordered breathing problems, and improves patients' sleep quality and health.

[0061] 2. In this invention, the apnea-hypopnea index is combined with multiple dimensions such as event duration factor, blood oxygenation decline factor, and event type factor to calculate a severity score, which is then classified into mild, moderate, and severe. This not only improves the accuracy and scientific rigor of assessing the severity of apnea events, but also allows for the selection of the optimal intervention from an intervention strategy library based on the real-time severity, severity level, and snoring type of the event, thus enabling personalized intervention plans. This allows for timely and effective responses to each patient's apnea events, minimizing disruption to the patient's sleep and improving their sleep quality and quality of life. Attached Figure Description

[0062] Figure 1 This is a flowchart of the sleep apnea monitoring system based on non-contact intervention of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] To address the issues of existing PSG methods requiring the attachment of numerous electrodes and sensors to the patient's body, which is not only cumbersome but also expensive; and the discomfort and disruption to natural sleep caused by home-use contact devices, please refer to [the relevant documentation / reference]. Figure 1 This embodiment provides the following technical solution:

[0065] Sleep apnea monitoring systems based on non-contact intervention include:

[0066] The non-contact signal acquisition unit is configured to collect biological signals containing respiratory waveform, heartbeat waveform and body movement information based on a biological radar sensor. The biological radar sensor can adopt an ultra-wideband radar with high resolution and anti-interference capability, and is arranged under a mattress or at the head of a bed. The snoring sound data and sleep environment data are monitored in real time based on a microphone array and a temperature and humidity sensor installed at the head of the bed.

[0067] The sleep data analysis unit is configured to collect historical records, extract biological feature parameters related to apnea from historical biological signals, including but not limited to respiratory amplitude, respiratory interval, respiratory rhythm disorder degree, heart rate variability and blood oxygen saturation, extract apnea and hypopnea events from the biological feature parameters based on a preset threshold as training data, and the sleep data analysis unit includes:

[0068] The threshold setting module is configured to select biological signals of multiple patients from historical biological signals, extract biological feature parameters related to apnea, and calculate the median of the biological feature parameters of each patient as an initial threshold. The multiple patients cover different genders, ages, BMIs and different severity of apnea to ensure the diversity of the data.

[0069] The biological signals and snoring sound data of the multiple patients within a specified unit of time without significant abnormalities are obtained as auxiliary feature parameters, and the baseline level of the auxiliary feature parameters of each patient is calculated as an auxiliary threshold.

[0070] The sleep environment data corresponding to the biological signals of the multiple patients are obtained as environmental factors. The environmental factors learn the general influence of environmental changes on biological feature parameters from historical biological signals, for example, the average respiratory amplitude baseline decreases by 2% for every 10% decrease in humidity.

[0071] The weighted adaptive threshold function is introduced to calculate the initial threshold, auxiliary threshold and environmental factors of the multiple patients to obtain the threshold confidence of the multiple patients. The threshold confidence of the multiple patients is sorted, and the patient with the highest threshold confidence is extracted, and the initial threshold of the patient is used as the optimal threshold for subsequent detection of apnea and hypopnea events.

[0072] The weighted adaptive threshold function is introduced to calculate the initial threshold, auxiliary threshold and environmental factors of the multiple patients to obtain the threshold confidence of the multiple patients, including:

[0073] The initial weight is obtained based on the initial threshold, auxiliary threshold and environmental factors of the historical influence degree of apnea.

[0074] The initial weight is optimized by using the weighted adaptive threshold function to obtain the optimized weight.

[0075] The initial threshold value, the auxiliary threshold value and the environmental factor are integrated with corresponding optimization weights to obtain the threshold confidence of each patient.

[0076] In this embodiment, the calculation formula of the threshold confidence is:

[0077]

[0078] wherein, is the threshold confidence of the i th patient, is the parameter value of the j th biological characteristic parameter of the i th patient, is the median of the j th biological characteristic parameter of the i th patient, is the allowable fluctuation range of the initial threshold value, is the initial threshold value of the patient, is the auxiliary threshold value of the patient, is the environmental influence factor, is the influence weight of the initial threshold value, is the influence weight of the auxiliary threshold value, is the influence weight of the environmental influence factor, wherein n is the number of biological characteristic parameters of the i th patient, the sum of the influence weights of the initial threshold value, the auxiliary threshold value and the environmental influence factor is 1, the influence weights of the initial threshold value, the auxiliary threshold value and the environmental influence factor are determined based on SHAP values, and the auxiliary threshold value is determined based on the baseline level of the auxiliary characteristic parameter of the patient.

[0079] The above-mentioned beneficial effects are achieved: by fusing the biological characteristic parameters, the auxiliary signal baseline and the environmental factor, the threshold confidence of the patient is determined, which can improve the accuracy of monitoring, considering the environmental factor can make the determined threshold confidence more environmentally adaptable, considering the auxiliary threshold value can make the stability performance of the threshold confidence better, thereby improving the anti-interference performance of the threshold confidence.

[0080] The sleep data analysis unit further comprises:

[0081] The fluctuation range of the threshold confidence of each patient in a state without sleep is obtained;

[0082] The threshold confidence of the patient whose fluctuation range is greater than a preset fluctuation threshold value is removed to obtain a first threshold confidence set;

[0083] The physiological characteristic data of each patient in the first threshold confidence set is obtained, and the physiological characteristic data includes the patient's disease data;

[0084] The comorbidity disease data related to sleep apnea is extracted from a preset disease database;

[0085] The disease data of each patient is compared with the comorbidity data, and the patient corresponding to the disease data matched with the comorbidity data is taken as a corrected patient;

[0086] The threshold confidence of the corrected patient is removed from the first threshold confidence set to obtain a second threshold confidence set;

[0087] The threshold confidences in the second threshold confidence set are sorted, the patient with the highest threshold confidence is extracted, and the initial threshold corresponding to the patient is taken as a first threshold;

[0088] The physiological feature data and the behavior habit data of the patient corresponding to the first threshold are obtained, the physiological feature data includes the patient's patient size, and the behavior habit data includes the patient's patient sleep posture;

[0089] Based on the deviation degree of the patient's patient size from the median of the standard size range interval, a first influence threshold of the patient is determined;

[0090] Based on the deviation degree of the patient's patient sleep posture from the standard sleep posture, a second influence threshold of the patient is determined;

[0091] The first threshold of the patient is threshold-optimized by using the first influence threshold and the second influence threshold to obtain a second threshold;

[0092] The sleep stage of the patient is determined based on the physiological feature data of the patient, and the second threshold is optimized in stages based on the different sleep stages to obtain an optimized optimal threshold, the optimal threshold includes an optimized threshold of each sleep stage.

[0093] In this embodiment, the state of not sleeping refers to the threshold confidence of the patient in the state of being awake or not sleeping, wherein the state of not sleeping includes but is not limited to the state of daytime sitting, meditation, reading, etc., and the amplitude of the change of the threshold confidence with time can reflect the threshold stability of the patient, and the threshold confidence with weak threshold stability cannot be taken as the optimal threshold.

[0094] In this embodiment, the preset fluctuation threshold refers to the maximum fluctuation amplitude of the threshold confidence allowed for screening stable patients, and the preset fluctuation threshold is determined according to the statistical results of all patient data, for example, if the fluctuation range of 95% of the patients is less than 0.2, the preset fluctuation threshold is 0.2, and if the fluctuation range of the patient is greater than 0.2, the threshold confidence of the patient is removed.

[0095] In this embodiment, the first threshold confidence set refers to the confidence set composed of the threshold confidences of the remaining patients after removing the patients with a threshold confidence fluctuation greater than the preset fluctuation threshold from the patients.

[0096] In this embodiment, the physiological characteristic data refers to the biological characteristics exhibited by the patient in daily life, covering morphological structure (such as height, skin color), functional performance (such as respiratory rate, hormone secretion rhythm), and metabolic activity (such as energy conversion, material synthesis), etc.

[0097] In this embodiment, the disease data refers to the disease data information that has been diagnosed in the patient, wherein the disease data includes disease name, diagnosis time, ICD code, etc.

[0098] In this embodiment, the preset disease database is a structured medical knowledge base, which contains information of other diseases that have clear pathophysiological association with the target disease (such as sleep apnea), and includes the name, ICD code, association strength, pathophysiological mechanism, etc. of different diseases in the preset disease database.

[0099] In this embodiment, the comorbidity data refers to the comorbidity list associated with sleep apnea in the preset disease database, for example, the comorbidity data includes data of hypertension, coronary heart disease, heart failure, atrial fibrillation, stroke, metabolic syndrome, depression, etc.

[0100] In this embodiment, the corrected patient refers to the patient whose disease data matches the comorbidity data in the preset disease database, that is, the patient who has other diseases that interfere with the patient's sleep apnea.

[0101] In this embodiment, the second threshold confidence set refers to the set of threshold confidence of the remaining patients after removing the threshold confidence of the corrected patients from the first threshold confidence set.

[0102] In this embodiment, the first threshold refers to the initial threshold corresponding to the patient with the highest threshold confidence in the second threshold confidence set, wherein the initial threshold generally includes apnea duration, etc.

[0103] In this embodiment, the patient's body type refers to the body shape characteristics quantified by BMI or body fat rate.

[0104] In this embodiment, the standard body type range is determined according to the preset standard, for example, the WHO standard: BMI 18.5~24.9 is normal, <18.5 is thin, and ≥25 is overweight, so the median of the standard range is 21.7.

[0105] In this embodiment, the sleep posture refers to the habitual sleeping position of the patient (supine, lateral, prone, etc.), wherein for patients with sleep apnea, lateral position has lower risk of sleep apnea due to tongue root falling back, so lateral position is the standard sleep posture.

[0106] In this embodiment, the first influence threshold is an adjustment coefficient determined based on the degree of deviation of the patient's body shape from the median of the standard body shape range. For example, the degree of deviation = / Standard median, first impact threshold= .

[0107] In this embodiment, the second influence threshold is an adjustment coefficient of the first threshold determined based on the degree of deviation of the patient's sleeping posture from the standard sleeping posture. For example, if the standard sleeping posture is side-lying, then the second influence threshold when the patient is lying on their back is 1.2, and the second influence threshold when lying on their side is 1.0.

[0108] In this embodiment, the second threshold refers to the optimized threshold obtained after considering the influence of the patient's body shape and posture. The second threshold is determined by multiplying the first threshold, the first influence threshold, and the second influence threshold. For example, if patient A's first threshold is 14 seconds (duration of apnea), the first influence threshold is 1.03, and the second influence threshold is 1.2, then the second threshold is 17.304 seconds.

[0109] In this embodiment, the sleep stages are defined as sleep cycle stages based on the patient's electroencephalogram (EEG). For example, sleep stages can be divided into light sleep and deep sleep, or into light sleep, moderate deep sleep, deep sleep, and REM sleep. Among these, the risk of sleep apnea is higher during deep sleep and REM sleep.

[0110] In this embodiment, the phased optimization threshold refers to optimizing the second threshold according to different sleep stages, and reducing the second threshold in sleep stages with a higher risk of sleep apnea.

[0111] In this embodiment, the optimal threshold refers to the set of thresholds obtained by adjusting the second threshold according to the sleep stage. For example, if the sleep stage is divided into four stages, the set of thresholds of the optimal threshold includes four or fewer threshold confidence levels.

[0112] The working principle of the above is as follows: First, the threshold confidence fluctuations of multiple patients in non-sleep states are initially screened for threshold confidence. Then, patients with comorbid sleep apnea are identified and removed, resulting in a second threshold confidence set. The threshold of the patient with the highest threshold confidence is extracted from the second threshold confidence set as the first threshold. The first threshold is then optimized based on the patient's body type and sleep posture. Finally, the first threshold is further optimized based on the patient's sleep stage, so that the output optimal threshold can meet the needs of each sleep stage, achieving more personalized optimization of the optimal threshold for sleep apnea.

[0113] The beneficial effects achieved by the above are: the threshold confidence of the patient is optimized multiple times through the stability of the threshold confidence, the comorbidity with other diseases, the body type of the patient, the sleep posture, and the sleep stage, so that the optimal threshold can adapt to the monitoring needs of the patient's sleep apnea in different physiological states, improve the monitoring adaptability, reduce the misjudgment rate, and make the determination of the patient's sleep apnea more accurate.

[0114] In one embodiment, assuming that there are three patients' biological characteristic parameter data, the average respiratory interval of patient 1 is 10 seconds, that of patient 2 is 12 seconds, and that of patient 3 is 11 seconds; the snoring baseline level of patient 1 is 30 dB, that of patient 2 is 35 dB, and that of patient 3 is 32 dB; the environmental factors of patient 1 are 22℃ and 50% humidity, those of patient 2 are 23℃ and 55% humidity, and those of patient 3 are 22.5℃ and 52% humidity; the weighted adaptive threshold function is applied to each patient, and the calculation results are that the threshold confidence of patient 1 is 28, that of patient 2 is 32, and that of patient 3 is 29; after sorting, it can be known that the threshold confidence of patient 2 is the highest, and the initial threshold of patient 2, i.e., the respiratory interval of 12 seconds, is taken as the optimal threshold for the subsequent detection of apnea and hypopnea events.

[0115] The data processing module is configured to preprocess the biological signal, including but not limited to filtering, noise reduction, and separation, to obtain a preprocessed biological signal, so as to ensure that the biological signal meets the requirements of the convolutional neural network model training.

[0116] The data division module is configured to filter out apnea and hypopnea events from the biological characteristic parameters based on a preset threshold, as a training set of the convolutional neural network model, for training the convolutional neural network model; and take the preprocessed biological signal as a test set, for verifying whether the prediction result of the convolutional neural network model is accurate.

[0117] The beneficial effects achieved by the above are: the initial threshold, the auxiliary threshold, and the environmental factors are comprehensively considered by the adaptive threshold function, the threshold confidence of each patient is calculated, and the initial threshold of the patient with the highest threshold confidence is selected as the optimal threshold; to improve the clinical practicability and reliability of the threshold, and then ensure the effectiveness of subsequent identification of biological characteristic parameters, and improve the accuracy and reliability of the detection of apnea and hypopnea events, so as to provide more accurate monitoring and diagnosis for the patient, help to discover and intervene sleep breathing disorders in time, and improve the sleep quality and health level of the patient.

[0118] The apnea recognition unit is configured to construct a convolutional neural network model including multiple convolutional layers, pooling layers, and fully connected layers, and learn biological feature parameters of apnea and hypopnea events in the data using the convolutional neural network model. Specifically, the convolutional layer extracts local features in the training data through convolution operations, such as specific patterns in the respiratory waveform, abnormal changes in the heart beat waveform, or specific actions in the body movement information. The pooling layer extracts global features through down-sampling operations, enhances the robustness of the features, and reduces the risk of overfitting of the model. The local and global features extracted through the fully connected layer are fused to form a comprehensive feature representation, and it is determined whether the current signal belongs to an apnea or hypopnea event.

[0119] The apnea intervention unit is configured to calculate the number of apnea and hypopnea events occurring in a unit time, and evaluate the severity accordingly, such as mild, moderate, and severe. When an apnea event is determined to occur, the corresponding intervention measure is automatically triggered in combination with the severity. The apnea intervention unit includes:

[0120] The evaluation standard setting module is configured to calculate the number of apnea and hypopnea events occurring in a unit time in the historical biological signals of multiple patients as an apnea-hypopnea index. The duration of each apnea and hypopnea event of each patient is recorded, and the average duration of all events is calculated as an event duration factor. For example, if the average duration of an event is 30 seconds, the factor is 1.0; if the average duration is 15 seconds, the factor is 0.5.

[0121] The oxygen saturation decrease trend of each patient is calculated by the biological radar sensor, and the average oxygen decrease amplitude of all events is calculated as an oxygen decrease factor. The snoring sound data of multiple patients is analyzed to classify the snoring sound data, and the classification includes obstructive and central, as an event type factor. The obstructive type is given a higher weight, such as 1.0, and the central type is given a lower weight, such as 0.7.

[0122] The weighted summation method is used to calculate the apnea-hypopnea index and the event duration factor, the oxygen decrease factor, and the event type factor, respectively, to obtain a severity score. According to the severity scores of multiple patients, they are divided into mild, moderate, and severe. For example, normal: CSS < 20; mild: 20 <= CSS < 40; moderate: 40 <= CSS < 70; severe: CSS >= 70; CSS represents the severity score.

[0123] The intervention module is configured to select an optimal intervention measure from an intervention strategy library when it is determined that the apnea event occurs, by combining the severity of the real-time event, the severity level, and the snoring category to which the event belongs. The intervention measures include, but are not limited to, inducing the subject to change the sleeping posture through a vibration device, stimulating the laryngeal muscle to expand the airway through an electric stimulation device, adjusting the hardness and shape of the mattress through an intelligent mattress to improve the breathing condition, and adjusting the indoor temperature, humidity, and ventilation through an environmental control device to optimize the sleep environment.

[0124] In one embodiment, for example: patient 1 belongs to mild and is obstructive snoring, then a first-level intervention is triggered, such as: slightly adjusting the neck angle through an intelligent pillow; patient 2 belongs to moderate and is obstructive snoring, then a second-level intervention is triggered, such as: on the basis of the first-level intervention, the head of the bed is slowly lifted by 10°; patient 3 belongs to severe and is accompanied by serious blood oxygen drop, then a third-level intervention is triggered, such as: on the basis of the second-level intervention, the environment light is slowly brightened for gentle stimulation, and the family members are notified.

[0125] The feedback module is configured to continue monitoring the breathing signal of the user through the bio-radar sensor after the intervention is implemented, to form a closed-loop feedback; if the breathing does not recover to normal within a set time, an escalation strategy is started, such as: increasing the lifting angle; if it has recovered to normal, the intervention is slowly withdrawn after a period of time, such as: lowering the head of the bed, so that the patient benefits without feeling, and the sleep interruption is minimized.

[0126] The above-mentioned beneficial effects are achieved: by combining the apnea hypopnea index with the event duration factor, the blood oxygen drop factor, and the event type factor in multiple dimensions, the severity score is calculated, and accordingly, it is divided into mild, moderate, and severe; not only the accuracy and scientificity of the judgment of the severity of the apnea event are improved, but also the optimal intervention measure is selected from the intervention strategy library according to the severity of the real-time event, the severity level, and the snoring category, to realize the individualized intervention scheme, so that the apnea event of each patient can be responded to in time and effectively, the interference to the sleep of the patient is minimized, and the sleep quality and the life quality of the patient are improved.

[0127] The report generation unit is configured to automatically generate a visual sleep health report based on the sleep data analysis results of the patient every night; the report content shows the night respiratory event distribution, the heart rate change, the blood oxygen trend, the intervention measure triggering record and the effect evaluation in the form of a time axis, and provides individualized health improvement suggestions based on the multi-day data trend analysis.

[0128] The user interaction unit is configured to provide an interactive interface for the family members and medical personnel of the patient through a mobile terminal application program, to display the current breathing waveform, heart beat waveform, body movement information, snoring data, and sleep environment data, as well as the historical sleep health report in real time, to realize the visualization of the monitoring data and intervention feedback.

[0129] An alarm unit is configured to automatically generate and push a hierarchical warning information to a preset guardian mobile phone application or medical platform when a high-risk event is detected, such as long-term severe apnea with a large decrease in blood oxygen or continuous multiple nights of excessive severity, while triggering intervention measures; the warning information includes an event summary and an operation button for requesting a response, such as received, ignored, and contact the user.

[0130] A recording unit is configured to record all alarm events and their intervention measure processing states after the implementation of intervention measures, facilitating subsequent user review and analysis.

[0131] A storage unit is configured to store biological signal data and corresponding event labels of a patient after long-term intervention treatment, forming a data set.

[0132] An optimization unit is configured to periodically use the data set to perform incremental learning and fine-tuning on the threshold of the biological feature parameter, the severity score, and the convolutional neural network model, to adapt to the physiological changes of the patient, and to achieve more accurate personalized monitoring and intervention effects.

[0133] A collaborative management unit is configured to support patient family members to adjust the sensitivity and notification method of intervention measures and to confirm, correct, or label events recorded by the system based on personal preferences through a mobile terminal application; for example, displaying the current triggered intervention measures and providing detailed explanations of the intervention measures; setting emergency contact information to promptly notify in the event of a severe apnea event.

[0134] Working principle: biological signals such as respiratory waveform, heart beat waveform, and body movement information of the patient are collected, and snoring sound and sleep environment data are monitored in real time; biological feature parameters related to apnea are identified from historical biological signals, and apnea and hypopnea events are screened out as training data based on preset thresholds; the convolutional neural network model learns the feature parameters of apnea and hypopnea events in the training data, thereby accurately identifying related events; and the severity of apnea is set according to the training results, and when an apnea event is detected, the corresponding intervention measures are automatically triggered according to the severity, to timely relieve the apnea condition.

[0135] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it should be taken in a descriptive sense and not a limiting sense.

[0136] While the embodiments of the application have been shown and described herein, it is to be understood that the application is not limited to these embodiments. Rather, it is to be understood that changes, modifications, substitutions and alterations can be made to the embodiments without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

Claims

1. A sleep apnea monitoring system based on non-contact intervention, characterized in that, include: The non-contact signal acquisition unit is configured to acquire the patient's biological signals based on a bio-radar sensor, and to monitor snoring data and sleep environment data in real time based on a microphone array and temperature and humidity sensors. The sleep data analysis unit is configured to collect historical records and extract biometric parameters related to sleep apnea from historical biosignals, including respiratory amplitude, respiratory interval, respiratory rhythm disorder, heart rate variability, and blood oxygen saturation; and extract sleep apnea and hypoventilation events from the biometric parameters based on preset thresholds as training data. The apnea recognition unit is configured to build a convolutional neural network model, use the convolutional neural network model to learn the biometric parameters of apnea and hypoventilation events in the training data, and identify apnea and hypoventilation events from them. The apnea intervention unit is configured to calculate the number of apnea and hypoventilation events occurring per unit time and assess their severity accordingly. When an apnea event is detected, appropriate intervention measures are automatically triggered based on the severity. The sleep data analysis unit includes: The threshold setting module is configured to select biosignals from multiple patients from historical biosignals, extract biosignature parameters related to apnea, and calculate the median of biosignature parameters for each patient as the initial threshold. Acquire biosignal and snoring data from multiple patients within a specified time period without significant abnormalities, as auxiliary feature parameters; calculate the baseline level of the auxiliary feature parameters for each patient, as an auxiliary threshold. Sleep environment data corresponding to the biosignals of multiple patients were acquired as environmental factors; a weighted adaptive threshold function was introduced to calculate the initial threshold, auxiliary threshold and environmental factors of multiple patients to obtain the threshold confidence of multiple patients. The threshold confidence scores of multiple patients were ranked, and the patient with the highest threshold confidence score was extracted. The initial threshold of the patient with the highest threshold confidence score was used as the optimal threshold for subsequent detection of apnea and hypoventilation events. The introduced weighted adaptive threshold function calculates the initial threshold, auxiliary threshold, and environmental factors for multiple patients to obtain the threshold confidence scores for multiple patients, including: Initial weights are obtained based on the historical influence of initial thresholds, auxiliary thresholds, and environmental factors on apnea; optimized weights are obtained by using a weighted adaptive threshold function; and the initial thresholds, auxiliary thresholds, and environmental factors are combined with the corresponding optimized weights to obtain the threshold confidence level for each patient. The sleep data analysis unit further includes: Obtain the fluctuation range of the threshold confidence score for each patient when they are not asleep; remove the threshold confidence scores of patients whose fluctuation range is greater than the preset fluctuation threshold to obtain the first threshold confidence score set; Obtain physiological characteristic data for each patient in the first threshold confidence set, wherein the physiological characteristic data includes the patient's disease data; Extract comorbidity data related to sleep apnea from a pre-defined symptom database; compare each patient's symptom data with the comorbidity data, and identify patients whose symptom data matches the comorbidity data as corrected patients; Remove the threshold confidence scores of corrected patients from the first threshold confidence score set to obtain the second threshold confidence score set; sort the threshold confidence scores in the second threshold confidence score set, extract the patient with the highest threshold confidence score, and use the initial threshold corresponding to the patient as the first threshold; Obtain physiological characteristic data and behavioral habit data of the patient corresponding to the first threshold. The physiological characteristic data includes the patient's body shape, and the behavioral habit data includes the patient's sleeping posture. Determine the first influence threshold of the patient based on the degree of deviation of the patient's body shape from the median of the standard body shape range. Based on the degree of deviation between the patient's sleep posture and the standard sleep posture, a second impact threshold for the patient is determined; the first impact threshold and the second impact threshold are used to optimize the first threshold to obtain the second threshold; The patient's sleep stage is determined based on the patient's physiological characteristic data, and the second threshold is optimized in stages based on different sleep stages to obtain the optimized optimal threshold, which includes the optimized threshold for each sleep stage.

2. The sleep apnea monitoring system based on non-contact intervention according to claim 1, characterized in that, The sleep apnea intervention unit includes: The assessment criteria setting module is configured to calculate the number of apnea and hypoventilation events that occurred in multiple patients per unit time in historical biosignals, as the apnea-hypopnea index. Record the duration of each apnea and hypoventilation event for each patient and calculate the average duration of all events as the event duration factor; The blood oxygen saturation decline trend of each patient is estimated by using bio-radar sensors, and the average blood oxygen decline rate of all events is calculated as a blood oxygen decline factor. We analyzed snoring data from multiple patients, categorized the snoring data into obstructive and central types, and assigned higher weights to obstructive snoring and lower weights to central snoring. Using a weighted summation method, the severity score was calculated by combining the apnea-hypopnea index with multiple dimensions, including event duration factor, blood oxygenation decline factor, and event type factor. Based on the severity scores of multiple patients, they were classified into mild, moderate, and severe.

3. The sleep apnea monitoring system based on non-contact intervention according to claim 2, characterized in that, The sleep apnea intervention unit also includes: The intervention module is configured to, when a sleep apnea event is detected, obtain the severity of the event in real time, the severity level, and the snoring category to which the event belongs, and combine these to select the optimal intervention measure from the intervention strategy library; The feedback module is configured to continue monitoring the user's breathing signals through a bio-radar sensor after the intervention is implemented, forming a closed-loop feedback. If breathing does not return to normal within a set time, an escalation strategy is initiated. If breathing has returned to normal, the intervention is slowly withdrawn after a period of time, so that the patient can benefit without feeling anything and minimize sleep interruption.

4. The sleep apnea monitoring system based on non-contact intervention according to claim 1, characterized in that, The sleep data analysis unit further includes: The data processing module is configured to preprocess biological signals, including but not limited to filtering, noise reduction and separation, to obtain preprocessed biological signals. The data partitioning module is configured to select apnea and hypoventilation events from biometric parameters based on preset thresholds, and use them as the training set for training the convolutional neural network model; the preprocessed biological signals are used as the test set to verify the accuracy of the prediction results of the convolutional neural network model.

5. The sleep apnea monitoring system based on non-contact intervention according to claim 3, characterized in that, Interventions include inducing subjects to change their sleeping position through vibration devices, stimulating laryngeal muscles to dilate the airway through electrical stimulation devices, adjusting the firmness and shape of the mattress through smart mattresses to improve breathing, and regulating indoor temperature, humidity and ventilation through environmental control devices to optimize the sleep environment.

6. The sleep apnea monitoring system based on non-contact intervention according to claim 1, characterized in that, Also includes: The report generation unit is configured to automatically generate a visual sleep health report based on the analysis results of the patient's sleep data each night; The user interaction unit is configured to provide an interactive interface for patients' families and medical personnel through a mobile terminal application, displaying in real time the current respiratory waveform, heart rate waveform, body movement information, snoring data, and sleep environment data, as well as historical sleep health reports, thereby enabling visualization of monitoring data and intervention feedback.

7. The sleep apnea monitoring system based on non-contact intervention according to claim 6, characterized in that, Also includes: The alarm unit is configured to automatically generate and push tiered alarm information to the preset guardian's mobile application or medical platform when a high-risk event is detected, while triggering intervention measures. The recording unit is configured to record all alarm events and their intervention status after intervention measures are implemented; The storage unit is configured to store the patient's biosignal data and corresponding event labels after long-term intervention and treatment, forming a dataset. The optimization unit is configured to periodically use the dataset to incrementally learn and fine-tune the thresholds, severity scores, and convolutional neural network models of biometric parameters to adapt to the patient's physiological changes. The collaborative management unit is configured to allow patients' families to adjust the sensitivity and notification methods of interventions according to their personal preferences, as well as to confirm, correct, or annotate events recorded by the system, based on a mobile terminal application.

Images