Method and apparatus for detecting respiratory disorders

Abstract

Methods and devices provide monitoring of a person's sleep disordered breathing state, such as for screening. One or more sensors can be configured for non-contact active and / or passive sensing. One or more processors (7304, 7006) can extract one or more breathing effort signals from one or more motion signals generated by active non-contact sensing using one or more sensors. One or more processors can extract one or more energy band signals from acoustic audio signals generated by passive non-contact sensing using one or more sensors. One or more processors can evaluate energy band signals and / or breathing effort signals to generate intensity signals representing sleep disordered breathing modulation. One or more processors can classify one or more features derived from one or more intensity signals to generate one or more sleep disordered breathing measures. The processor can generate a sleep disordered breathing indication based on one or more sleep disordered breathing measures. Some versions can evaluate one or more sensing signals to generate one or more indications of one or more cough events and / or cough types.

Description

[0001] 1 Cross-reference to related applications

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 62 / 769,272, filed on November 19, 2018, and U.S. Provisional Patent Application No. 62 / 769,272, filed on December 23, 2019, the entire contents of which are incorporated herein by reference.

[0003] 2 STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0004] not applicable

[0005] 3 Names of the parties involved in the joint research and development

[0006] not applicable

[0007] 4. Sequence Listing

[0008] not applicable 5 Background Technology 5.1 Technical Field

[0011] The present technology relates to one or more of the detection, diagnosis, treatment, prevention and improvement of respiratory disorders such as sleep apnea and / or cough. The present technology also relates to medical devices or equipment and their uses. In particular, the present technology relates to screening devices or equipment and their uses, such as non-contact screening devices.

[0012] 5.2 Description of related technologies

[0013] 5.2.1 Human respiratory system and its disorders

[0014] The human respiratory system facilitates gas exchange. The nose and mouth form the entrance to the patient's airway.

[0015] The airways include a series of branching tubes that become narrower, shorter, and more numerous as they penetrate deeper into the lungs. The main function of the lungs is gas exchange, allowing oxygen to enter the venous blood from the air and expel carbon dioxide. The trachea is divided into the left main bronchi and the right main bronchi, which are ultimately divided into terminal bronchioles. The bronchi constitute the conducting airways, but do not participate in gas exchange. Further branches of the airways lead to the respiratory bronchioles and ultimately to the alveoli. The alveolar region of the lungs is the area where gas exchange occurs and is called the respiratory zone. See Respiratory Physiology, 9th edition, by John B. West, published by Lippincott Williams & Wilkins in 2011.

[0016] There is a range of breathing disorders. Some disorders can be characterised by specific events such as apnea, hypopnea and hyperpnea.

[0017] Examples of breathing disorders include obstructive sleep apnea (OSA), Cheyne-Stokes respiration (CSR), respiratory insufficiency, obesity hyperventilation syndrome (OHS), chronic obstructive pulmonary disease (COPD), neuromuscular disease (NMD), and chest wall disorders.

[0018] Obstructive sleep apnea (OSA) is a form of sleep-disordered breathing (SDB) characterized by events including occlusion or obstruction of the upper airway during sleep. It is caused by abnormal narrowing of the upper airway during sleep combined with a normal decrease in muscle tone in the area of ​​the tongue, soft palate, and posterior oropharyngeal wall. The condition causes affected patients to stop breathing, typically for a period of 30 to 120 seconds, sometimes 200 to 300 times per night. This often leads to excessive daytime sleepiness and can lead to cardiovascular disease and brain damage. This complication is a common disorder, particularly in middle-aged overweight men, although the affected person may not be aware of the problem. See U.S. Patent No. 4,944,310 (Sullivan).

[0019] Cheyne-Stokes respiration (CSR) is another form of sleep-disordered breathing. CSR is a disorder of the patient's respiratory controller in which there are rhythmic alternating cycles of boom-and-bust ventilation called CSR cycles. CSR is characterized by repeated hypoxia and reoxygenation of arterial blood. CSR can be harmful because of the repeated lack of oxygen. In some patients, CSR is associated with repeated arousals from sleep, which leads to severe sleep disruption, increased sympathetic activity, and increased afterload. See U.S. Patent No. 6,532,959 (Berthon-Jones).

[0020] Obesity hyperventilation syndrome (OHS) is defined as severe obesity combined with chronic hypercapnia during wakefulness in the absence of other known causes of hypoventilation. Symptoms include dyspnea, morning headache, and excessive daytime sleepiness.

[0021] Chronic obstructive pulmonary disease (COPD) includes any of a group of lower airway diseases with certain common features. These diseases include increased resistance to air flow, prolonged expiratory phase of breathing, and loss of normal elasticity of the lungs. Examples of COPD are emphysema and chronic bronchitis. COPD is caused by chronic smoking (main risk factor), occupational exposure, air pollution, and genetic factors. Symptoms include: exertional dyspnea, chronic cough, and sputum production. COPD is the fourth leading cause of death worldwide, with more than 65 million people classified as moderate or severe, affecting up to a quarter of adults over 40 years old (World Health Organization http: / / www.who.int / respiratory / copd / burden / en / ).

[0022] Neuromuscular disease (NMD) is a broad term that covers many diseases and ailments that impair muscle function directly through intrinsic muscle pathology or indirectly through neuropathology. Some NMD patients are characterized by progressive muscle damage, which leads to loss of walking ability, wheelchair use, dysphagia, respiratory muscle weakness, and ultimately death from respiratory failure. Neuromuscular disorders can be divided into rapidly progressive and slowly progressive: (i) rapidly progressive disorders: characterized by muscle damage that worsens over months and leads to death within a few years (e.g., amyotrophic lateral sclerosis (ALS) and Duchenne muscular dystrophy (DMD) in adolescents; (ii) variable or slowly progressive disorders: characterized by muscle damage that worsens over years and only slightly shortens life expectancy (e.g., limb-girdle, facioscapulohumeral, and myotonic muscular dystrophy). Symptoms of respiratory failure in NMD include: increasing general weakness, dysphagia, dyspnea during exercise and at rest, fatigue, drowsiness, morning headaches, and difficulty concentrating and mood changes.

[0023] Chest wall disorders are a group of chest deformities that result in inefficient connections between the respiratory muscles and the thorax. These disorders are usually characterized by restrictive defects and have the potential for long-term hypercapnic respiratory failure. Scoliosis and / or kyphoscoliosis can cause severe respiratory failure. Symptoms of respiratory failure include: dyspnea on exertion, peripheral edema, orthopnea, recurrent chest infections, morning headaches, fatigue, poor sleep quality, and loss of appetite.

[0024] Heart failure (HF) is a relatively common and serious clinical condition characterized by the inability of the heart to keep up with the body's oxygen demand. Due to its high prevalence and severity, the management of heart failure is a major challenge facing modern healthcare systems. HF is a chronic condition that is progressive in nature. The progression of HF is usually characterized by relative stability over a long period of time (although cardiovascular function is reduced), with occasional acute attacks. In these acute attacks, patients experience worsening symptoms, such as dyspnea (dyspnea, cholecystokinetic rhythm, increased jugular venous pressure, and orthopnea. This is often accompanied by significant congestion (which is the accumulation of fluid in the lung cavity). This excess fluid often results in a significant increase in weight of several kilograms. However, in many cases, when significant congestion occurs, doctors have limited options to help patients regain stability, and in many cases, patients require hospitalization. In extreme cases, if not treated promptly, patients may experience an acute decompensated heart failure (ADHF) event, sometimes referred to as decompensation.

[0025] 5.2.2 Treatment

[0026] Various therapies, such as continuous positive airway pressure (CPAP) therapy, non-invasive ventilation (NIV), invasive ventilation (IV), and high flow therapy (HFT) have been used to treat one or more of the above-mentioned breathing disorders.

[0027] Continuous positive airway pressure (CPAP) therapy has been used to treat obstructive sleep apnea (OSA). The mechanism of action is that the continuous positive airway pressure acts as a pneumatic splint and can prevent upper airway occlusion, such as by pushing the soft palate and tongue forward and away from the posterior oropharyngeal wall. Treatment of OSA by CPAP therapy can be voluntary, so if the patient finds the device used to provide such treatment to be any one or more of: uncomfortable, difficult to use, expensive, and unsightly, the patient may choose not to comply with the treatment.

[0028] Non-invasive ventilation (NIV) provides ventilation support to the patient through the upper airway to help the patient breathe and / or maintain appropriate oxygen levels in the body by completing some or all of the work of breathing. Ventilatory support is provided via a non-invasive patient interface. NIV has been used to treat CSR and respiratory failure in the form of OHS, COPD, NMD and chest wall disorders. In some forms, the comfort and effectiveness of these treatments can be improved.

[0029] Invasive ventilation (IV) provides ventilation support to patients who are unable to breathe effectively on their own and can be provided using a tracheostomy tube. In some forms, the comfort and effectiveness of these treatments can be improved.

[0030] 5.2.3 Treatment system

[0031] The above-mentioned treatment can be provided by the treatment system or device. The system and device can also be used to diagnose the disease without treating it.

[0032] A therapy system, such as a respiratory therapy device, may include a respiratory pressure therapy device (RPT device), an air circuit, a humidifier, a patient interface, and data management.

[0033] 5.2.3.1 Patient interface

[0034] The patient interface can be used to engage the respiratory apparatus to its user, for example by providing an air flow. The air flow can be provided to the user's nose and / or mouth via a mask, to the mouth via a tube, or to the patient's trachea via a tracheostomy tube. Depending on the treatment to be applied, the patient interface can form a seal with, for example, the patient's facial area, thereby facilitating the delivery of gas at a pressure sufficiently different from the ambient pressure (e.g., a positive pressure of about 10 cmH2O) to achieve the treatment. For other forms of treatment, such as oxygen delivery, the patient interface may not include a seal sufficient to facilitate the delivery of a gas supply at a positive pressure of about 10 cmH2O to the airway.

[0035] 5.2.3.2RPT device

[0036] Air pressure generators are known in a range of applications, such as industrial-scale ventilation systems. However, medical air pressure generators have specific requirements that are not met by more general air pressure generators, such as reliability, size and weight requirements for medical devices. Furthermore, even devices designed for medical use may have disadvantages with respect to one or more of the following: comfort, noise, ease of use, efficacy, size, weight, manufacturability, cost and reliability.

[0037] 5.2.3.3 Humidifier

[0038] Delivering an air stream without humidification may result in airway drying. Using a humidifier with an RPT device and a patient interface to produce humidified gas minimizes drying of the nasal mucosa and improves patient airway comfort. Additionally, in colder climates, warm air, which is often applied to the facial area in and around the patient interface, is more comfortable than cold air.

[0039] 5.2.4 Diagnostic and monitoring systems

[0040] Diagnosis is the identification of a condition based on its signs and symptoms. Diagnosis tends to be a one-time process, while monitoring the progression of symptoms can continue indefinitely. Some diagnostic systems are only suitable for diagnosis, while some can also be used for monitoring. Screening generally involves a diagnostic process such as monitoring signs and systems over time that can be evaluated relative to the specific condition being diagnosed.

[0041] Of interest is the ability to detect SDB and other sleep disorders. Sleep-disordered breathing can be considered abnormal breathing during sleep, such as breathing stops or "sleep apnea." The most common is obstructive sleep apnea, in which the upper airway collapses despite continued efforts. Other types include central sleep apnea with lack of breathing and lack of effort, and mixed apnea. As a sleep-related condition, screening for diagnosis or even monitoring can be a difficult task. Sleep apnea is over 80% undiagnosed (Peppard PE et al. Am J Epidemiol 2013; Young T et al. Sleep 1997). For example, for every 4 people with sleep apnea, there are 22 who are unaware that they have the condition. Therefore, a simple, readily available means of screening for sleep-disordered breathing or sleep apnea to detect the condition could have significant societal benefits.

[0042] Polysomnography (PSG) is a conventional system for monitoring patients. PSG typically involves placing 15 to 20 contact sensors on a person to record various body signals, such as electroencephalogram (EEG), electrocardiogram (ECG), electrooculogram (EOG), etc. Such a large number of contact sensors and leads are prone to failure, expensive, and have limited accessibility, making PSG an undesirable method for detecting sleep-related disorders.

[0043] The S+ (pronounced ess-plus) (ResMed Sensor Technologies Ltd, Dublin, Ireland) is a contactless bedside monitor suitable for long-term monitoring of chronic diseases such as HF and COPD. The S+ contains a bio-motion transceiver sensor operating at ultra-low power (less than 1 mW) in the unlicensed bands of 5.8 GHz or 10.5 GHz. The S+ is capable of measuring body movement at distances ranging from 0.3 to 1.5 meters; in the case of two occupants in the bed, a combination of sophisticated sensor design and intelligent signal processing enables the S+ to measure movement of only the person closest to the sensor. The S+ is suitable for long-term monitoring of SDB because it is unobtrusive and there are no significant compliance issues. However, processing the raw S+ signal to obtain information useful for SDB screening, monitoring, or diagnosis is a difficult and complex task. 6 Summary of the invention

[0045] The present technology is intended to provide a medical device for diagnosing, monitoring, ameliorating, treating or preventing breathing disorders, such as sleep apnea and / or cough, having one or more of improved comfort, cost, efficacy, ease of use and manufacturability.

[0046] One aspect of the present technology relates to devices for screening, diagnosing or monitoring sleep breathing disorders (including, for example, different types of apneas, hypopneas, flow limitation, snoring) and / or coughing.

[0047] Another aspect of the present technology relates to a method for screening, diagnosing, or monitoring sleep-disordered breathing and / or cough using a non-contact sensor.

[0048] Some versions of the present technology may include a method of one or more processors for monitoring a person's sleep disordered breathing state. The method in the one or more processors may include extracting one or more breathing signals from one or more motion signals, the one or more motion signals being generated by active contactless sensing; the method in the one or more processors may include extracting one or more energy band signals from passive signals, the passive signals being generated by passive contactless sensing. The passive signal may represent acoustic information detected by a sound sensor. The method in the one or more processors may include evaluating one or more energy band signals and / or one or more breathing signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation. The method in the one or more processors may include classifying one or more features derived from the one or more intensity signals to generate one or more sleep disordered breathing measures. Any of classifying features and generating sleep disordered breathing measures may be remotely generated or processed by one or more remote processing devices or servers. Therefore, in some embodiments, the method may include transmitting data for classification or classified features for generating sleep disordered breathing measures to a remote device or server.

[0049] In some versions, the method in the one or more processors may include generating a sleep disordered breathing indication based on one or more sleep disordered breathing measures. The method in the one or more processors may include displaying on a display or controlling display and / or forwarding a sleep disordered breathing indication. The method in the one or more processors may include displaying on a display or controlling display and / or forwarding one or more sleep disordered breathing measures. One or more breathing signals may include a breathing effort signal. One or more sleep disordered breathing measures may include a probability of sleep disordered breathing. Classification may include identifying one of a positive and negative number of sleep disordered breathing events during a sleep period that exceeds a threshold, and wherein the one or more sleep disordered breathing measures may include a binary flag indicating the result of the identification. When the probability exceeds the threshold, the binary flag may indicate a positive. Sleep disordered breathing events may include at least one of apnea events and hypopnea events. One or more sleep disordered breathing measures may include an apnea-hypopnea index that represents an estimate of the total number of apnea events and hypopnea events.

[0050] In some versions, the method in the one or more processors may include generating a sleep stage adjustment factor based on probability. The method in the one or more processors may include adjusting the sleep stage time according to the adjustment factor. The method in the one or more processors may include generating a cluster flag signal based on one or more intensity signals. The cluster flag signal may represent a time series that identifies whether SDB modulation exists. The cluster flag signal may be generated based on a comparison between the value of one or more intensity signals and a threshold. When the value of one or more intensity signals is greater than a first intensity threshold, the flag of the cluster flag signal may be set to true. The cluster flag signal may be further set based on an evaluation of the value of the filtered signal compared to a second intensity threshold. The filtered signal may be derived by filtering one or more intensity signals. One or more features may include one or more total sleep time proportions with SDB clusters. One or more features may include peak intensity or peak average intensity.

[0051] In some versions, the method in the one or more processors may include generating a sleep-wake correction mask signal based on a generated cluster marker signal characterizing the presence of a detected SDB cluster. The method in the one or more processors may include applying the sleep-wake correction mask signal to a sleep segmentation process, in which wake classification instances are corrected to sleep instances based on the sleep-wake correction mask signal. The generated sleep disordered breathing indication may include a graphical risk-o-meter displayed on a display device, the graphical risk-o-meter including a pointer and a scale. The scale may be represented using an indication of a discrete range of sleep disordered breathing risk. Evaluating one or more respiratory signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation includes any one, more, or all of the following: (a) generating an envelope signal; (b) normalizing the envelope signal; and (c) generating a spectral characteristic from the normalized envelope signal. The spectral characteristic may include a peak frequency of a power spectral density operating within a sleep disordered breathing frequency range. The spectral characteristics may include an in-band metric that may include a ratio of: (a) power in a narrowband near the peak frequency or peak frequency of a power spectral density operation, and (b) power in a sleep disordered breathing frequency range from the power spectral density operation. The spectral characteristics may include an in-band metric that may include a ratio of: (a) power in a narrowband near the peak frequency of a power spectral density operation, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in a narrowband near the peak frequency. The in-band metric may be an average metric derived from in-band metric values ​​from the I channel motion signal and the Q channel motion signal.

[0052] In some versions, evaluating one or more respiratory signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation includes any, more, or all of the following: (a) combining one or more energy band signals; (b) generating an envelope signal from the combined energy band signals; (c) filtering and normalizing the envelope signal from the combined energy band signals; and (d) generating a spectral characteristic from the filtered and normalized envelope signal. The spectral characteristic from the filtered and normalized envelope signal may include an in-band metric including a ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep disordered breathing frequency range from the power spectral density operation. The spectral characteristic from the filtered and normalized envelope signal may include an in-band metric including a ratio of: (a) power in a narrow band near the peak frequency of the power spectral density operation, and (b) a difference between the total power of the spectrum of the power spectral density operation and the power in a narrow band near the peak frequency.

[0053] In some versions, extracting one or more breathing signals from one or more motion signals may include combining multiple motion signals. Each of the multiple motion signals may be a motion signal representing motion from a detection range that is different from the detection range of other motion signals in the multiple motion signals. The combination may include calculating a weight for each of the multiple motion signals based on the breathing frequency from the power spectral density, and determining a weighted average of the absolute values ​​of the multiple motion signals. Extracting one or more energy band signals from a passive signal may include any one, more, or all of the following: (a) separating the sound frequencies of the passive signal into band signals by calculating a transform of the passive signal; (b) calculating energy values ​​of the band signals; and (c) calculating an average of the calculated energy values ​​for each band signal.

[0054] In some versions, active contactless sensing may include sonar sensing using a microphone and a speaker. Active contactless sensing may include radar sensing using a radio frequency transmitter and a receiver. Active contactless sensing may include frequency modulated continuous wave (FMCW) sensing. Passive contactless sensing may include acoustic sensing of breathing-related sounds using a microphone. The method in the one or more processors may include preprocessing the sound signal generated by the microphone to produce a passive signal. The preprocessing includes one, more, or all of the following: (a) filtering using an infinite impulse response filter; (b) removing a baseline, including subtracting a minimum value over a sliding window; (c) removing artifacts, using percentile limits; (d) normalizing with a standard deviation over a sliding window; and (e) integrating and high-pass filtering. The method in the one or more processors may include autocorrelating the preprocessed sound signal. The method in the one or more processors may include detecting a peak of a predefined breathing range of the autocorrelated, preprocessed sound signal. The method in the one or more processors may include determining a respiratory rate estimate based on peaks of multiple signals. Each of the plurality of signals may be a sound signal of a discrete frequency band and processed by preprocessing and autocorrelation. The one or more features may be a plurality of features derived from a plurality of intensity signals from the one or more energy band signals and the one or more respiratory signals, and the generated one or more sleep disordered breathing measures may be generated by classifying the plurality of features.

[0055] In some versions, the one or more processors may be in a processing device, which may include any of a smartphone, a tablet computer, a general purpose computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device. The method in the one or more processors may include controlling, using the one or more processors, changes to a therapy setting of the respiratory therapy device based on one or more sleep disordered breathing measures.

[0056] In some versions where at least some of the processing is performed on one or more remote devices, the method may also include, in one or more processors, receiving back from the remote device the generated one or more sleep disordered breathing measures. The method may also include (a) displaying the received one or more sleep disordered breathing measures on a display; or (b) transmitting the received one or more sleep disordered breathing measures to a local processing / display device via a data communication transmission.

[0057] Some versions of the present technology may include a processor-readable medium having processor-executable instructions stored thereon, which, when executed by one or more processors, cause the one or more processors to monitor a person's sleep disordered breathing state, the processor-executable instructions being configured to perform any of the methods described herein.

[0058] Some versions of the technology may include a server having access to a processor-readable medium as claimed in claim 1, wherein the server may be configured to receive a request to download processor-executable instructions of the processor-readable medium to a processing device over a network.

[0059] Some versions of the present technology may include a processing device including: one or more processors; a speaker coupled to the one or more processors; a microphone coupled to the one or more processors; and a processor-readable medium as described herein. The processing device may include any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

[0060] Some versions of the present technology may include a processing device including one or more processors; a microphone coupled to the one or more processors; a radio frequency sensor coupled to the one or more processors; and a processor-readable medium as described herein.

[0061] Some versions of the present technology may include a method for a server having access to a processor-readable medium as described herein. The method may include receiving at the server a request to download processor-executable instructions of the processor-readable medium to a processing device over a network. The method may include transmitting the processor-executable instructions to the processing device in response to the request.

[0062] Some versions of the present technology may include a device for monitoring a person's sleep disordered breathing state. The device may include one or more processors. The one or more processors may be configured to extract one or more breathing signals from one or more motion signals, and the one or more motion signals are generated by active non-contact sensing using one or more sensors. One or more processors may be configured to extract one or more energy band signals from passive signals, and the passive signals are generated by passive non-contact sensing using one or more sensors. The passive signal may represent acoustic information detected by a sound sensor in one or more sensors. One or more processors may be configured to evaluate one or more energy band signals and / or one or more breathing signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation. One or more processors may be configured to classify or transmit one or more features derived from one or more intensity signals for classification to generate one or more sleep disordered breathing measures. One or more processors may also be configured to generate sleep disordered breathing indications based on one or more sleep disordered breathing measures.

[0063] In some versions, the device may include one or more sensors, one or more sensors configured for active contactless sensing and passive contactless sensing. Alternatively, the device may be configured to communicate with external sensors (e.g., smart speakers, etc.) that have been positioned in the surrounding environment of the device and utilize data from external sensors. One or more processors may be coupled to one or more sensors in communication. One or more sensors may be configured for active contactless sensing and passive contactless sensing. One or more processors may also be configured to display and / or forward sleep disordered breathing indications on a display. One or more processors may also be configured to display and / or forward one or more sleep disordered breathing measures on a display. One or more breathing signals may include a breathing effort signal. One or more sleep disordered breathing measures may include the probability of sleep disordered breathing. Classification may include identifying that the number of sleep disordered breathing events during a sleep period exceeds a threshold value, one of affirmation and negation. One or more sleep disordered breathing measures may include a binary flag representing the result of the identification. When the probability exceeds a threshold value, the binary flag may indicate affirmation. Sleep disordered breathing events may include at least one of apnea events and hypopnea events. One or more sleep disordered breathing measures may include an apnea-hypopnea index representing an estimate of the total number of apnea events and hypopnea events. One or more processors may also be configured to generate a sleep stage adjustment factor based on probability. One or more processors may also be configured to adjust the sleep stage time according to the adjustment factor. One or more processors may also be configured to generate a cluster flag signal based on one or more intensity signals. The cluster flag signal may represent a time series that identifies whether SDB modulation exists. The cluster flag signal may be generated based on a comparison between the value of one or more intensity signals and a threshold. When the value of one or more intensity signals is greater than a first intensity threshold, the flag of the cluster flag signal may be set to true. The cluster flag signal may be further set based on an evaluation of the value of the filtered signal compared to the second intensity signal. The filtered signal may be derived by filtering one or more intensity signals. One or more features may include one or more total sleep time proportions with SDB clusters. One or more features may include peak intensity or peak average intensity.

[0064] In some versions, one or more processors may also be configured to generate a sleep-wake correction mask signal based on a generated cluster marker signal characterizing the presence of a detected SDB cluster. One or more processors may also be configured to apply the sleep-wake correction mask signal to a sleep segmentation process, in which wake classification instances are corrected to sleep instances according to the sleep-wake correction mask signal. The generated sleep disordered breathing indication may include a graphical risk-o-meter displayed on a display device. The graphical risk-o-meter may include a pointer and a scale. The scale may be represented using an indication of a discrete range of sleep disordered breathing risk. In order to evaluate one or more respiratory signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation, one or more processors are configured to or control any, multiple, or all of the following: (a) generating an envelope signal; (b) normalizing the envelope signal; and (c) generating a spectral characteristic from the normalized envelope signal. The spectral characteristic may include a peak frequency of a power spectral density operating within a sleep disordered breathing frequency range. The spectral characteristics may include an in-band metric that may include a ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep disordered breathing frequency range from the power spectral density operation. The spectral characteristics may include an in-band metric that may include a ratio of: (a) power in a narrowband near the peak frequency of the power spectral density operation, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in a narrowband near the peak frequency. The in-band metric may be an average metric derived from in-band metric values ​​from the I channel motion signal and the Q channel motion signal.

[0065] In some versions, in order to evaluate one or more energy band signals to generate one or more intensity signals representing the intensity of sleep disordered breathing modulation, one or more processors may be configured, controlled or performed any, more or all of the following: (a) combining one or more energy band signals; (b) generating an envelope signal from the combined energy band signals; (c) filtering and normalizing the envelope signal from the combined energy band signals; and (d) generating spectral characteristics from the filtered and normalized envelope signal. The spectral characteristics from the filtered and normalized envelope signal may include an in-band metric including a ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep disordered breathing frequency range from the power spectral density operation. The spectral characteristics from the filtered and normalized envelope signal may include an in-band metric including a ratio of: (a) power in a narrow band near the peak frequency of the power spectral density operation, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in a narrow band near the peak frequency.

[0066] In some versions, to extract one or more respiratory effort signals from one or more motion signals, one or more processors may be configured to combine multiple motion signals, each of the multiple motion signals being a motion signal representing motion from a detection range that is different from the detection range of other motion signals in the multiple motion signals. To combine multiple motion signals, one or more processors may be configured to calculate a weight for each of the multiple motion signals based on a respiratory frequency from a power spectral density, and determine a weighted average of the absolute values ​​of the multiple motion signals. To extract one or more energy band signals from a passive signal, one or more processors are configured to control or execute any, multiple, or all of the following: (a) separating the sound frequencies of the passive signal into band signals by calculating a transform of the passive signal; (b) calculating energy values ​​of the band signals; and (c) calculating an average of the calculated energy values ​​for each band signal.

[0067] In some versions of the device, active contactless sensing may include sonar sensing, wherein one or more sensors may include a microphone and a speaker. Active contactless sensing may include radar sensing, wherein one or more sensors may include a radio frequency transmitter and a receiver. Active contactless sensing may include frequency modulated continuous wave (FMCW) sensing. Passive contactless sensing may include acoustic sensing of breathing-related sounds, wherein one or more sensors may include a microphone. One or more processors may be configured to preprocess the sound signal generated by the microphone to produce a passive signal. Preprocessing may include any, more, or all of the following: filtering using an infinite impulse response filter; removing a baseline, including subtracting a minimum value over a sliding window; removing artifacts, using percentile limits; normalizing with a standard deviation over a sliding window; and integrating and high-pass filtering. One or more processors may be configured to perform autocorrelation processing on the preprocessed sound signal. One or more processors may be configured to detect peaks of a predefined breathing range of the autocorrelation, preprocessed sound signal. One or more processors may be configured to determine a respiratory rate estimate based on peak values ​​of a plurality of signals, each of the plurality of signals being a sound signal in a discrete frequency band and processed by preprocessing and autocorrelation. One or more features may be a plurality of features derived from a plurality of intensity signals from one or more energy band signals and one or more effort breathing signals, and the generated one or more sleep disordered breathing measures may be generated by classifying the plurality of features. One or more processors may be in a processing device, the processing device comprising any of a general purpose computing device, a smart phone, a tablet computer, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device. One or more processors may be configured to control changes to a therapeutic setting of a respiratory therapy device based on one or more sleep disordered breathing measures. The device may include a processor-readable medium as described herein.

[0068] In some versions of the device, the one or more processors may also be configured to receive one or more sleep disordered breathing measures that have been generated and derived by processing by one or more external devices. The one or more processors may also be configured to (a) display the received one or more sleep disordered breathing measures on a display; or (b) transmit the received one or more sleep disordered breathing measures to a local processing / display device via a data communication transmission.

[0069] Some versions of the present technology may include a method of one or more processors for identifying a cough of a person. The method in the one or more processors may include accessing an acoustic signal generated using a microphone. The acoustic signal is generated by passive non-contact sensing in the vicinity of the person. The method in the one or more processors may include deriving one or more cough-related features from the acoustic signal. The method in the one or more processors may include classifying or transmitting for classification the one or more cough-related features to generate an indication of one or more cough events of the person.

[0070] In some versions, one or more cough-related features relate to whether there is a person's cough. Classifying one or more features may include identifying cough types. Cough types may include any one or more of the following: (a) dry cough type, (b) expectorant cough type, (c) wheezing-related cough type, and (d) spasm-related cough type. Classifying one or more features may include identifying cough attribution types. Cough attribution types may include any one or more of the following: (a) asthma cough type, (b) chronic obstructive pulmonary (COPD) cough type, (c) bronchitis cough type, (d) tuberculosis (TB) cough type, (e) pneumonia cough type, (f) lung cancer cough type, (g) gastroesophageal reflux disease (GERD), and (h) upper airway cough syndrome. One or more processors are also configured to generate a cough intensity measure indicating the intensity level of an event in one or more cough events. The cough intensity measure may include an acoustic amplitude value and / or a loudness value. One or more processors may also be configured to determine the variability of the cough intensity measure. The one or more features derived from the acoustic signal may include any, more, or all of the following: frequency features, temporal features, spectrogram features, and wavelet features.

[0071] In some versions, frequency-dependent features of the one or more features derived from the acoustic signal may include any one, more, or all of the following: (1) local peaks, (2) ratios of a main peak to one or more surrounding peaks, (3) local maxima, (4) global maxima; (5) harmonics, (6) integrals of one or more frequency components, (8) ratios of different frequency energy estimates, (7) one or more Mel-frequency cepstral coefficients (MFCCs), (9) spectral flux, (10) spectral centroid, (11) harmonic product spectrum, (12) spectral stretch, (13) one or more spectral autocorrelation coefficients, (14) spectral kurtosis, and (15) linear predictive coding (LPC). Time-dependent features of the one or more features derived from the acoustic signal may include any one, more, or all of the following: (1) root mean square (RMS) value, (2) zero crossing rate, (3) envelope; and (4) pitch period based on an autocorrelation function. The method may include processing the acoustic signal through voice activity detection to suppress background noise in the acoustic signal. The method may also include estimating a cough rate from the acoustic signal. The method may also include estimating a change in the cough rate. One or more processors may be configured to extract respiratory features from the detected respiratory waveform. Classifying the one or more features to generate an indication of one or more cough events of a person may be based on the one or more respiratory features extracted from the detected respiratory waveform. The one or more respiratory features may include one, more, or all of the following: (1) inhalation time, (2) inhalation depth, (3) exhalation time, (4) exhalation depth, (5) inhalation-exhalation ratio, (6) one or more gaps in the respiratory waveform caused by coughing, and (7) respiratory rate. The one or more respiratory features may be derived using one or more of passive contactless sensing and active contactless sensing; the one or more processors may be configured to generate one or more motion signals by active contactless sensing using an active contactless sensing device. The one or more processors may be configured to generate an indication of one or more cough events of a person based on an evaluation of the generated one or more motion signals.

[0072] In some forms, the method may include detecting a person's body position. Evaluation of the generated one or more motion signals may include detecting biometrics unique to the person. Evaluation of the generated one or more motion signals may include detecting sleep stage information from the one or more motion signals. One or more processors may be configured to negate acoustically sensed cough events based on the detection of sleep stage information. One or more processors may be configured to attribute acoustically sensed cough events to a person based on the detection of sleep stage information. Active contactless sensing may include one or more of acoustic type sensing, optical type sensing, and radar type sensing. One or more processors may be configured to convey data indicating one or more cough events of a person to suggest further investigation of the condition and / or control of one or more of the following: environmental parameters, settings of a processing device, behavioral changes, and / or processing parameters; one or more processors may also be configured to generate reminders to change or wash bedding.

[0073] In some versions, the classification involves a classifier derived from any one or more of supervised machine learning, deep learning, a convolutional neural network, and a recurrent neural network. The method may also include monitoring sound to detect user environment interactions. The user environment interactions may include detecting user environment interaction features, which may include any one or more of clickers, appliances, and doors. Monitoring sound to detect user environment interactions may include evaluating activity patterns of the monitored person to generate an indication that contact with the monitored person is needed.

[0074] The methods, systems, devices, and apparatus described herein can improve functionality in a processor, such as a processor of a processing device (e.g., a dedicated computer), and / or a sleep and / or sleep disorder breathing diagnostic / monitoring device. In addition, in some cases, they can communicate with or be integrated into a controller or processor of a therapeutic device, such as a respiratory therapy device. In addition, the methods, systems, devices, and apparatus described can provide improvements in the art of automated management, screening, monitoring, and / or treatment of sleep-related conditions and / or respiratory conditions, including, for example, sleep disordered breathing.

[0075] Some versions of the present technology may include a processor-readable medium having processor-executable instructions stored thereon, which, when executed by one or more processors, cause the one or more processors to recognize a person's cough according to any of the methods described herein. Some versions of the present technology may include a server capable of accessing such a processor-readable medium, which receives a request to download the processor-executable instructions of the processor-readable medium to a processing device over a network. Such a processing device may include: one or more processors; a speaker, the speaker coupled to one or more processors; a microphone, the microphone coupled to one or more processors; and the processor-readable medium or one or more processors as described herein may be configured to access the processor-readable instructions using the server. The processing device may include any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device. The processing device may include one or more processors; a microphone, the microphone coupled to one or more processors; a radio frequency sensor, the radio frequency sensor coupled to one or more processors; and such a processor-readable medium.

[0076] Some versions of the present technology may include a method for a server capable of accessing such a processor-readable medium. The method may include receiving at the server a request to download processor-executable instructions of the processor-readable medium to a processing device over a network; and transmitting the processor-executable instructions to the processing device in response to the request.

[0077] Some versions of the present technology may include a processing device for identifying a person's cough. The device may include one or more microphones, one or more microphones configured for passive contactless sensing, wherein the one or more microphones generate acoustic signals by performing passive contactless sensing near a person. The device may include one or more processors coupled to the one or more microphones. The one or more processors may include a module configured to access an acoustic signal generated using one or more microphones. The one or more processors may include a module configured to derive one or more features from the acoustic signal. One or more features may relate to whether a person coughs. The one or more processors may include a module configured to classify or transmit one or more features for classification to generate an indication of one or more cough events of a person. The processing device may include any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device. The processor may also be configured to display the generated event indication on a display of the device, or forward the generated event indication to an external processing / display device.

[0078] The methods, systems, devices, and apparatus described herein can improve functionality in a processor, such as a processor of a processing device (e.g., a dedicated computer), and / or a sleep and / or sleep disorder breathing diagnostic / monitoring device. In addition, in some cases, they can communicate with or be integrated into a controller or processor of a therapeutic device, such as a respiratory therapy device. In addition, the methods, systems, devices, and apparatus described can provide improvements in the art of automated management, screening, monitoring, and / or treatment of sleep-related conditions and / or respiratory conditions, including, for example, sleep disordered breathing.

[0079] Of course, some parts of these aspects can form sub-aspects of the present technology. In addition, various sub-aspects and / or aspects can be combined in various ways, and also constitute additional aspects or sub-aspects of the present technology.

[0080] Other features of the present technology will be apparent by considering the following detailed description, abstract, drawings, and information contained in the claims. 7. Description of the drawings

[0082] The present technology is illustrated by way of example and not limitation in the figures of the accompanying drawings in which like reference numerals indicate similar elements, including:

[0083] 7.1 Treatment System

[0084] Figure 1 An exemplary treatment system according to one form of the present technology is shown. A patient 1000 wearing a patient interface 3000 receives a supply of air under positive pressure from an RPT device 4000. Air from the RPT device 4000 is humidified in a humidifier 5000 and delivered to the patient 1000 along an air circuit 4170.

[0085] 7.2 Respiratory system and facial anatomy

[0086] Figure 2 A schematic diagram of the human respiratory system including the nasal and oral cavities, larynx, vocal cords, esophagus, trachea, bronchi, lungs, alveolar sacs, heart, and diaphragm is shown.

[0087] 7.3 Patient Interface

[0088] Figure 3 A patient interface in the form of a nasal mask is shown in accordance with one form of the present technology.

[0089] 7.4RPT device

[0090] Figure 4A One form of RPT device according to the present technology is shown.

[0091] Figure 4B A form of the present technology is shown Figure 4ASchematic diagram of the pneumatic path of the RPT device. The upstream and downstream directions are indicated.

[0092] 7.5 Humidifier

[0093] Figure 5A and 5B An isometric view of a humidifier according to one aspect of the present technology is shown.

[0094] 7.6 Respiratory Waveform

[0095] Fig. 6A A model typical breathing waveform of a sleeping person is shown.

[0096] Figure 6B Polysomnographic data is shown for a patient's non-REM sleep breathing period, typically over about 90 seconds.

[0097] Figure 6C Polysomnographic data of an OSA patient are shown.

[0098] Fig.6D Shown are flow data for a patient while the patient was experiencing a series of complete obstructive apneas.

[0099] Fig. 6E Polysomnographic data of a patient with Cheyne-Stokes breathing including apnea are shown.

[0100] Fig. 6F Polysomnographic data of a patient including Cheyne-Stokes breathing with hypopnea is shown.

[0101] 7.7 Monitoring Equipment

[0102] Fig. 7A A sleeping patient is shown during a polysomnogram.

[0103] Figure 7B-1 An exemplary processing device for monitoring a sleeping person, for example for acoustic sensing, is shown, which may be suitable for implementing the processes of the present technology for diagnosing, monitoring, or screening for SDB.

[0104] Figure 7B-2 For example Figure 7B-1 Schematic diagram of an exemplary processing device of a processing device.

[0105] Figure 7B-2 is a conceptual diagram of a mobile computing device or processing device configured in accordance with some forms of the present technology (eg, using the sensing and / or SDB screening techniques described herein).

[0106] Figure 7C-1 An exemplary device for monitoring a sleeping patient, eg, using radio frequency sensing, in accordance with one form of the present technology is shown.

[0107] Figure 7C-2 is shown in more detail Figure 7C-1 Block diagram of the monitoring device.

[0108] Figure 8 Exemplary processes or modules are shown for use in some versions of the present technology for a processing device, for example, for processing sensed signals to provide an output indication of a sleep disordered breathing condition.

[0109] Fig. 8A Exemplary processes or modules are shown for use in some versions of the present technology for a processing device, for example, for processing sensed signals to provide output indications of coughing symptoms, events, and / or types.

[0110] Figure 8B Additional exemplary processes or modules are shown for use in some versions of the present technology for a processing device, for example, for processing sensed signals to provide an output indication of a coughing condition, event, and / or type.

[0111] Figure 8C An exemplary process or module for detecting characteristic patterns of snoring, nasal congestion, coughing, or breathing difficulties using the passive and / or active sensing techniques described herein is shown, for example, for implementing Fig. 8A or Figure 8B The cough assessment system shown in .

[0112] Fig.8D Exemplary processes or modules for processing active and passive sensing streams are shown, which may be suitable for use with, for example, Figure 8C The illustrated techniques are implemented; particular emphasis is placed on exemplary processes or modules for processing passive sensing streams.

[0113] Fig. 9 Various motion signals associated with breathing effort in different range bins are shown, representing motion at different sensing distances from the sensor, such as by implementing range gating using one of the motion sensing devices of the present technology.

[0114] Fig.10 A process for signal processing motion signals associated with different distances to extract information related to respiratory effort to produce a respiratory effort signal in some versions of the present technology is shown.

[0115] Fig.11

[0046] Signal processing of a respiratory effort signal by spectral analysis is shown for identifying active clusters associated with SDB in some versions of the present technique.

[0116] Fig.12Shown is a spectrogram of an exemplary acoustic signal generated by a sound sensor (eg, microphone) using passive sound detection such as implemented by a processing device as described herein.

[0117] Fig.13 A process for signal processing an acoustic signal, such as an acoustic signal generated by passive sound detection using a sound sensor, is shown in some versions of the present technology for identifying passive clusters associated with SDBs.

[0118] Fig.14 A graphical SDB risk identifier that may be generated in some versions of the present technology is shown that illustrates a normal assessment of SDB risk and an estimated count of apnea and / or hypopnea events.

[0119] Fig.15 A graphical SDB risk identifier that can be generated in some versions of the present technology is shown, which shows an at-risk sleeper assessment of SDB risk and an estimated count of apnea and / or hypopnea events. The meter shows a medium risk with a calculated probability in an exemplary range of 15% to 30%.

[0120] Fig.16 A graphical SDB risk identifier that can be generated in some versions of the present technology is shown, which shows an at-risk sleeper assessment of SDB risk and an estimated count of apnea and / or hypopnea events. The meter shows a high risk with a calculated probability in an exemplary range of 30% to 100%.

[0121] Fig.17 A graphical SDB risk identifier that may be generated in some versions of the present technology is shown that illustrates an at-risk sleeper assessment of SDB risk when not enough data or sensed signals are detected.

[0122] Figures 18 to 32 A graphical display on a common time scale of signal generation using active and passive sensing techniques described herein is shown with exemplary respiratory events (SDBs) detectable by the methods described herein.

[0123] Fig.33 An exemplary user interface for a display of a device described herein is shown that outputs SDB risk, snoring, and sleep quality indicators derived from signals of the active and / or passive sensing techniques described herein.

[0124] Fig.34 A similar method to that described herein is shown. Fig.33An exemplary user interface for a display of a device that outputs SDB risk (dangerous sleep), snoring, and sleep quality indicators derived from signals utilizing active and / or passive sensing techniques described herein.

[0125] Fig.35 An exemplary user interface for a display of a device described herein is shown that outputs a signal-derived snoring analysis utilizing the active and / or passive sensing techniques described herein.

[0126] Fig.36 An exemplary user interface for a display of a device described herein is shown that outputs a signal-derived sleep analysis (including a hypnogram and sleep session summary) utilizing the active and / or passive sensing techniques described herein.

[0127] Fig.37 An exemplary user interface for a display of a device described herein is shown in which sleep analysis and respiration rate displays derived from signals utilizing active and / or passive sensing techniques described herein are output on a common time scale.

[0128] Fig.38 A display with the respiration rate in the signal varying over time is shown. 8. Specific implementation methods

[0130] Before describing the present technology in further detail, it should be understood that the present technology is not limited to the specific examples described herein, which may be varied. It should also be understood that the terminology used in the present disclosure is only for the purpose of describing the specific examples described herein and is not intended to be limiting.

[0131] The devices and methods described below are particularly suitable for monitoring, for example for screening cough and / or sleep breathing disorders, and are described in those terms. However, the devices and methods described may also be applied to monitoring other events that affect a patient's breathing.

[0132] The methods described herein may implement SDB risk detection and / or cough detection using passive sensing techniques (sound monitoring) and / or one or more active sensing techniques (e.g., radar and / or sonar sensing), and may be implemented by one or more processors, such as any of the devices described in this specification. For example, the methods described in this specification may be performed by one or more processors, such as (i) using an application on a processing or computing device, such as a mobile phone (e.g., a smart phone), a smart speaker, a smart TV, a smart watch, or a tablet computer), which may be configured with a speaker and a microphone, such that the processing of the present application implements a synergistic fusion of passive sensing (sound monitoring associated with acoustic breathing) and active acoustic sensing (e.g., sonar using ultrasonic sensing). (ii) on a dedicated hardware device implemented as a radio frequency (RF) sensor (e.g., radar) and a microphone, (iii) on a dedicated hardware setup implemented as an RF sensor without audio; and / or (iv) on a dedicated hardware device implemented as an active acoustic sensing device without RF sensing. Other combinations of such devices will be recognized with respect to the details disclosed below. Thus, such devices may work independently or collaboratively to implement acoustic and RF based sensing utilizing any of the detection / monitoring methods described herein.

[0133] For example, a device having one or more processors for implementing the detection and / or display described herein can be integrated with one or more active and / or passive sensing hardware in the housing of the device. Optionally, in order to receive information associated with active and / or passive sensing, a processing device having one or more processors to implement the detection and / or display described herein can be separately coupled to external active and / or passive sensing hardware, for example, via wired and / or wireless communications. Such external sensors can be, for example, other sensing devices located in a room (e.g., a bedroom) where sensing occurs, and can be separate from, for example, a processing device (e.g., an RPT device, a smart phone, or other smart device) that implements the detection and / or display described herein.

[0134] In addition, one or more processors configured to implement its detection and / or display can work independently or collaboratively, such as through data communication (e.g., through wired and / or wireless transmission). For example, the processing device receiving the sensing signal can perform signal processing and detection thereby locally or remotely, or some combination of the two. In one such instance, the processing device (e.g., smart phone) can be configured to perform processing and detection (from the received sensing signal) locally (i.e., on the processing device) and display the result. In another such instance, the processing device (e.g., smart phone) can be configured to perform some processing and / or detection (from the received sensing signal) locally (i.e., on the processing device), and transmit data to a remote processing device (e.g., one or more servers) so that some processing and / or detection can be performed remotely. In some such cases, the remote processing device can transmit the result back to the local processing device to display on the local processing device. In other cases, such as where the local processing device is an RPT device with limited processing and / or display capabilities, the local processing device may transmit data received from the remote device to another local device (e.g., a smart phone) for further processing and / or display.

[0135] In another exemplary configuration, the processing device (e.g., a smart phone) can be configured to transmit data (e.g., sensing signals) to a remote processing device (e.g., one or more servers) so that processing and detection can be performed remotely. In this case, the remote processing device can transmit the results back to the local processing device for display on the local processing device or on another device associated with the local processing device.

[0136] 8.1 Treatment System

[0137] In one form, the present technology includes an apparatus or device for treating sleep disorders. The apparatus or device may include a respiratory pressure therapy (RPT) device 4000 for supplying pressurized air to a patient 1000 via an air circuit 4170 leading to a patient interface 3000 .

[0138] 8.1.1 Patient interface

[0139] A non-invasive patient interface 3000 according to one aspect of the present technology may include the following functional aspects: a seal-forming structure 3100, an inflatable chamber 3200, a positioning and stabilizing structure 3300, a vent 3400, a connection port 3600 for connecting to a form of air circuit 4170, and a forehead support 3700. In some forms, the functional aspects may be provided by one or more physical components. In some forms, one physical component may provide one or more functional aspects. In use, the seal-forming structure 3100 is arranged to surround the entrance of the patient's airway to facilitate the supply of positive pressure air to the airway. In some forms, a patient interface for delivering high flow therapy, for example, may be provided without a significant seal-forming structure.

[0140] 8.1.2 RPT device

[0141] According to one aspect of the present technology, the RPT device 4000 or respiratory therapy device includes mechanical, pneumatic and / or electrical components and is configured to execute one or more algorithms. The RPT device 4000 can be configured to generate an air flow for delivery to the airway of a patient, such as for treating one or more respiratory conditions described elsewhere in this document.

[0142] In one form, the RPT device 4000 is constructed and arranged to deliver air flow in the range of -20 L / min to +150 L / min while maintaining a positive pressure of at least 6 cmH2O, or at least 10 cmH2O, or at least 20 cmH2O. Thus, such a device may be configured as a positive airway pressure device or a high flow therapy device.

[0143] The RPT device can have an outer housing 4010 that is constructed in two parts: an upper portion 4012 and a lower portion 4014. In addition, the outer housing 4010 can include one or more panels 4015. The RPT device 4000 includes a chassis 4016 that supports one or more internal components of the RPT device 4000. The RPT device 4000 can include a handle 4018.

[0144] The pneumatic path of the RPT device 4000 can include one or more air path items, such as an inlet air filter 4112, an inlet muffler 4122, a pressure generator 4140 capable of supplying positive pressure air (e.g., a blower 4142), an outlet muffler 4124, and one or more converters 4270, such as a pressure sensor 4272 and a flow sensor 4274.

[0145] One or more air path items may be located within a removable separate structure that will be referred to as a pneumatic block 4020. The pneumatic block 4020 may be located within the outer housing 4010. In one form, the pneumatic block 4020 is supported by or formed as part of the chassis 4016.

[0146] The RPT device 4000 may have a power source 4210, one or more input devices 4220, a central controller 4230, a treatment device controller 4240, a pressure generator 4140, one or more protection circuits 4250, a memory 4260, a converter 4270, a data communication interface 4280, and one or more output devices 4290. The electrical components 4200 may be mounted on a single printed circuit board assembly (PCBA) 4202. In an alternative form, the RPT device 4000 may include more than one PCBA 4202. When implemented with the acoustic and / or radio frequency sensing components described herein, or when communicating with a device having such sensing components to receive one or more such sensing signals, having such a central controller 4230, having one or more processors, the RPT device may be a processing device as described in more detail herein.

[0147] 8.1.3 Air circuit

[0148] The air circuit 4170 according to one aspect of the present technology is a conduit or tube that is constructed and arranged to allow air flow to flow between two components, such as the RPT device 4000 and the patient interface 3000, in use.

[0149] 8.1.4 Humidifier

[0150] In one form of the present technology, a humidifier 5000 (e.g., Figure 5A ) to change the absolute humidity of the air or gas for delivery to the patient relative to the ambient air. Typically, the humidifier 5000 is used to increase the absolute humidity of the air flow and increase the temperature of the air flow (relative to the ambient air) before delivery to the patient's airway.

[0151] The humidifier 5000 may include a humidifier reservoir 5110, a humidifier inlet 5002 for receiving an air flow, and a humidifier outlet 5004 for delivering a humidified air flow. Figure 5A and Figure 5B As shown, the inlet and outlet of the humidifier reservoir 5110 can be a humidifier inlet 5002 and a humidifier outlet 5004, respectively. The humidifier 5000 can also include a humidifier base 5006, which can be adapted to receive the humidifier reservoir 5110 and include a heating element 5240.

[0152] 8.2 Respiratory Waveform

[0153] Fig. 6A A typical breathing waveform of a model while a person is sleeping is shown. The horizontal axis is time and the vertical axis is respiratory flow. Although the parameter values ​​can vary, a typical breath can have the following approximate values: tidal volume, Vt, 0.5L, inspiratory time, Ti, 1.6s, peak inspiratory flow, Q 峰值 , 0.4L / s, expiratory time, Te, 2.4s, peak expiratory flow, Q 峰值 , -0.5L / s. The total duration of breathing is T 总 About 4 seconds. A person usually breathes at a rate of about 15 breaths per minute (BPM) with a ventilation volume of about 7.5 L / min. Typical duty cycles, Ti and T 总 The ratio is about 40%.

[0154] Figure 6B Selected polysomnographic channels (pulse oximetry, flow, chest movement, and abdominal movement) of a patient during non-REM sleep breathing are shown, typically for a period of about 90 seconds, about 34 breaths. The top channel shows blood oxygen saturation (SpO2), with a scale ranging from 90% to 99% saturation in the vertical direction. The patient maintained about 95% saturation throughout the period shown. The second channel shows quantitative respiratory flow, with a scale ranging from -1 to +1 LPS in the vertical direction, and inspiration is positive. Chest movement and abdominal movement are shown in the third and fourth channels.

[0155] 8.3 Diagnostic, monitoring and / or screening systems

[0156] 8.3.1 Polysomnography

[0157] Fig. 7A A patient 1000 undergoing a polysomnogram (PSG) is shown. The PSG system includes a head box 2000 that receives and records signals from the following sensors: EOG electrodes 2015; EEG electrodes 2020; ECG electrodes 2025; submental EMG electrodes 2030; snoring sensor 2035; respiratory inductive plethysmogram (respiratory effort sensor) 2040 located on a chest strap; respiratory inductive plethysmogram (respiratory effort sensor) 2045 located on an abdominal strap; oral and nasal cannula 2050 with oral thermistor; photoplethysmograph (pulse oximeter) 2055; and body position sensor 2060. The electrical signals are referenced to a ground electrode (ISOG) 2010 located at the center of the forehead.

[0158] Although PSG systems have good quality and reliability, they are not well suited for long-term continuous monitoring and have limited mobility, which can cause irritation, pain, and discomfort to the monitored patients. These limitations have created a stronger demand for improved sleep monitoring systems.

[0159] 8.3.2 Unobtrusive Monitoring Equipment

[0160] The present technology is particularly related to systems, methods and devices for detecting the motion of a subject, the motion of the subject including, for example, respiratory motion and / or heart-related chest motion when the subject is asleep. Based on such breathing and / or other motion detection, the sleep state and sleep disordered breathing events (e.g., apnea, hypopnea, etc.) of the subject can be detected. For example, a processing device, such as an application (e.g., software of one or more processors) associated with a processing device (e.g., a general computing device, a smart phone, a tablet computer, a smart speaker, a smart TV, a smart watch, etc.), can use a mobile device sensor, such as a speaker and a microphone and / or a radio frequency transmitter / receiver, to passively detect such audible sound and / or actively detect motion in a non-contact manner, for example. When combined together, such passive and active non-contact sensing technologies can complement each other and have synergistic benefits in improving sleep disordered breathing classification, even if these methods are used alone and may be confusing or incomplete. For the screening method described herein, such sensing signal information can be processed in a non-contact manner, wherein the sensor can monitor the user from a distance without directly or indirectly contacting the sensing modality of the subject user. As understood herein, a general purpose computing device may be any electronic device having a processor and access to a speaker and a microphone, such as for implementing any of the processing methods described herein.

[0161] A. Acoustic non-contact sensing equipment

[0162] Reference now Figure 7B-1 , Figure 7B-2 and Figure 7B-3, describes an exemplary device for implementing a system suitable for the present technology. A mobile phone, such as a processing device 7100, or an electronic device, configured with an application 7200 for detecting movement of a subject 1000 can be placed on a bedside table near a person or subject (e.g., patient 1000). The mobile phone or processing device 7100 can be, for example, a smart phone or tablet computer having one or more processors. Among them, the processor can be configured to execute the functions of the application 7200. Therefore, in process 7202, such functions can be included (a)-passive sensing, such as by using a sound transducer such as a microphone to generate an audio signal of audible ambient sound sensed in the environment of the device, and / or (b)-active sensing, such as by generating and transmitting an acoustic sensing signal using a speaker in a manner similar to sonar (SONAR), typically through air, which is a generally open or unrestricted medium in a room near the device, for example, sensing the transmitted signal using a transducer such as a microphone to receive reflections of the transmitted signal, and processing the sensed signal to determine body movement as one or more motion signals. In this document, the term sonar may be understood to refer to active acoustic sensing, for example by generating / transmitting ultrasound or low-frequency ultrasound sensing signals (e.g., in the frequency range of about 17 to 23 kHz, 18 to 22 kHz, or 17 to 18 kHz) through the air. Such systems may be considered with respect to PCT / EP201 7 / 073613, the disclosure of which is incorporated herein by reference.

[0163] Therefore, the processing device 7100 may include a speaker and a microphone in addition to other components. The speaker may be activated to transmit a sensing audio signal, and the microphone may be activated to receive a reflected signal or an ambient sound (eg, a user's breathing sound, etc.).

[0164] Therefore, according to some aspects of the present technology, an audio signal can be generated and sent to a user using one or more tones, for example. The tone provides a pressure change in a medium (e.g., air) at a specific frequency. For the purpose of this specification, the generated tones (or audio signals or sound signals) can be referred to as "sounds", "acoustics" or "audio" because they can be generated in a similar manner to audible pressure waves (e.g., by a speaker). However, such pressure changes and tones should be understood herein as being audible or inaudible, although they are characterized by any of the terms "sound", "acoustics" or "audio". Therefore, the generated audio signal can be audible or inaudible, wherein the audible frequency threshold in the crowd varies with age. The typical "audio" standard range is about 20Hz to 20,000Hz (20kHz). The high-frequency hearing threshold decreases with age, and middle-aged people usually cannot hear sounds above 15 to 17kHz, while teenagers can hear 18kHz. The most important frequencies of speech are approximately in the range of 250 to 6,000Hz. The speaker and microphone signal responses for typical consumer smartphones are designed to roll off above 19 to 20kHz in many cases, with some extending to above 23kHz and higher (especially where the device supports sampling rates greater than 48kHz, such as 96kHz). Therefore, for most people, signals in the range of 17 / 18 to 24kHz can be used and remain inaudible. For young people who can hear 18kHz but not 19kHz, a frequency band of 19kHz to 21kHz can be used. It should be noted that some household pets may be able to hear higher frequencies (e.g., dogs up to 60kHz, cats up to 79kHz). The technologies that have been referred to as "active" and "passive" sensing in this specification generally correspond to specific frequency ranges. For example, "active" sonar sensing is silent. Therefore, "active" sensing tends to focus on frequencies that are difficult or impossible to hear, which are about and above 18kHz. On the other hand, "passive" sensing targets audible sounds emitted by sleeping people and therefore generally focuses on the frequency range below 18kHz, such as 15Hz to 17kHz or even 250Hz to 6,000Hz.

[0165] The audio signal to be transmitted for active sensing (e.g., sonar) may include, for example, a sinusoidal waveform, a sawtooth chirp, a triangular chirp, etc. As background, as used herein, the term "chirp" is a short-term non-stationary signal, which may, for example, have a sawtooth or triangular shape, with a linear or nonlinear profile. Several types of signal processing methods may be used to generate and sense audio signals, including, for example, continuous wave (CW) homodyne detection, pulsed CW homodyne detection, frequency modulated CW (FMCW), frequency hopping range gating (FHRG), adaptive FHRG (AFHRG), ultra-wideband (UWB), orthogonal frequency division multiplexing (OFDM), adaptive CW, frequency shift keying (FSK), phase shift keying (PSK), binary phase shift keying (BSPK), quadrature phase shift keying (QPSK), and a generalized QPSK called quadrature amplitude modulation (QAM), etc.

[0166] Other motion signals generated by sonar-type contactless sensing devices may be used by examples of such apparatus, such as described in International PCT Patent Publication No. WO2018 / 050913 (PCT / EP2017 / 073613), filed on September 19, 2017, entitled "APPARATUS, SYSTEM, AND METHOD FOR DETECTING PHYSIOLOGICAL MOVEMENT FROM AUDIO AND MULTIMODAL SIGNALS", the entire contents of which are incorporated herein by reference. For example, such sonar sensing may enable substantially simultaneous, parallel sensing of multiple ranges from the sensor to generate multiple motion signals having motions from different ranges of the sensor, such as by generating FMCW sensing signals.

[0167] In general, and as described in more detail herein, the application 7200 can then process these detected signals to determine / extract relevant information (e.g., motion) from the active sensing signals and / or the acoustic respiratory characteristics from the passive sensing signals at process 7204. By evaluating the extracted information at process 7206, sleep-related information (e.g., identified apneas and / or hypopneas, and quantified sleep-disordered breathing) can be characterized based on such processing. Then, at process 7208, one or more identifiers can be generated and output.

[0168] Other processes for such an apparatus or such an apparatus may include those described in U.S. Provisional Patent Application Nos. 62 / 610,033, 62 / 610,013, and 62 / 609,998, each filed on December 22, 2017, the disclosures of which are incorporated herein by cross-reference.

[0169] like Figure 7B-3As shown, such processing device 7100 may include an integrated chip, memory and / or other control instructions, data or information storage medium. For example, programming instructions including the evaluation / signal processing method described herein may be encoded on an integrated chip in the memory of a device or equipment to form an application-specific integrated chip (ASIC). Such instructions may also or alternatively be loaded as software or firmware using an appropriate data storage medium. Optionally, such processing instructions may be downloaded from a server to a mobile device, for example, via a network (e.g., the Internet), so that when the instructions are executed, the processing device is used as a screening or monitoring device.

[0170] Therefore, the processing device 7100 may include Figure 7B-3 The processing device 7100 may optionally include any of a microphone or sound sensor 7302, a processor 7304, a display interface 7306, a user control / input interface 7308, a speaker 7310, and a memory / data storage 7312, as well as other components, such as processing instructions with the processing methods / modules described herein. Although not shown, such a processing device may also include a communication interface, such as for receiving or sending information such as sensor data from an external sensor.

[0171] One or more components of the processing device 7100 may be integrated with or operably coupled to the processing device 7100. For example, the microphone or sound sensor 7302 may be integrated with the processing device 7100 or coupled to the processing device 7100, for example, via a wired or wireless link (e.g., Bluetooth, Wi-Fi, or other communication interface, etc.). For example, a radio frequency sensing device as described herein may be coupled to the processing device 7100 to provide a sensing signal from the radio frequency sensing device to the processing device (or vice versa) for use in the processing described herein.

[0172] The memory / data storage 312 may include a plurality of processor control instructions for controlling the processor 304 of the processing device. For example, the memory / data storage 312 may include processor control instructions for causing the application 200 to be executed by the processing instructions of the processing method / module described herein. Optionally, such processor control instructions (codes) may be loaded as software or firmware using an appropriate data storage medium or processor readable medium. Optionally, such processing instructions may be downloaded from a server to a processing device or a mobile computing device, for example, via a network (e.g., the Internet or the Internet), so that when the instructions are executed, the processing device is used as a screening, diagnostic or monitoring device. Therefore, the server may be configured to respond to a request from a processing device, for example, by sending a processor control instruction (code) to the processing device via a network. Such a server may be configured to receive a request to download a processor executable instruction from a processor readable medium of a server to a processor readable medium of one or more processing devices via a network.

[0173] B. RF contactless sensing equipment

[0174] refer to Figure 7C-1 , other sensing devices can be considered, Figure 7C-1 An unobtrusive monitoring device 7000 in accordance with one form of the present technology is shown. The monitoring device 7000 is positioned adjacent and relatively close to a sleeping patient 1000 (e.g., on a bedside table), and may optionally communicate with or be integrated with a processing device as previously described.

[0175] Figure 7C-2 A more detailed description of a form of the present technology is shown in Figure 7C-1 1000. In the monitoring device 7000, the non-contact sensor unit 1200 includes a non-contact motion sensor 7010 generally directed toward the patient 1000. The motion sensor 7010 is configured to generate one or more signals representing body movements of the patient 1000, from which one or more respiratory motion signals representing the respiratory motion of the patient can be derived.

[0176] The sensor unit 1200 may also include a microcontroller unit (MCU) 7001, and a memory 7002 (e.g., a memory card) for recording data. In one implementation, the sensor unit 1200 may include a communication circuit 7004 configured to transmit data to an external computing device 7005 (e.g., a local general-purpose computer such as the processing device 7100) or a remote server via a connection 7008. The connection 7008 may be wired or wireless, in which case the communication circuit 7004 has wireless capabilities, and may be direct or indirect via a local area network or a wide area network (not shown).

[0177] The sensor unit 1200 includes a processor 7006 configured to process signals generated by the motion sensor 7010, as described in detail below.

[0178] The sensor unit 1200 may optionally include a display device 7015 configured to provide visual feedback to the user. In one implementation, the display device 7015 includes one or more warning lights (e.g., one or more light emitting diodes, etc.). The display device 7015 may also be implemented as a display screen such as an LCD or a touch-sensitive display. The processor 7006 controls the operation of the display device 7015 based on the assessment of the patient's sleep. The display device 7015 may be operated to display information to a user (e.g., patient 1000) or a physician or other clinician of the monitoring device 7000. The display device 7015 may also display a graphical user interface for operating the monitoring device 7000.

[0179] Sensor unit 1200 may also include an audio output 7017 configured to provide acoustic feedback to the user under the control of processor 7006, such as a tone whose frequency varies with breathing, or an alarm that sounds when certain conditions are met. In some versions, sensor unit 1200 may include an acoustic transducer, such as a microphone for passive sensing, as described in more detail herein.

[0180] User control of the operation of monitoring device 7000 may be based on control operations (not shown) sensed by processor 7006 of monitoring device 7000 .

[0181] It should be noted that, as indicated by the dashed line surrounding the motion sensor 7010, this sensor may be defined and positioned spatially separately from the rest of the components of the sensor unit 1200. In such a configuration, the contactless motion sensor 7010 may be arranged to communicate wirelessly with the rest of the unit.

[0182] One example of a sensor unit 1200 is the S+ device manufactured by ResMed Sensor Technologies Ltd, which may include a non-contact Doppler radio frequency (RF) motion sensor 7010. Such a sensor unit may be the sensor described in International PCT Patent Publication No. WO 2014 / 015238 filed on July 13, 2013, the entire disclosure of which is incorporated herein by reference.

[0183] In one form of the present technology, for example when using the S+ device, the sensor unit 1200 of the motion sensor 7010 includes an RF transmitter 7020 configured to transmit a signal 7060. Contrary to the acoustic non-contact sensing device discussed above, in this case the transmitted signal 7060 is in the radio frequency (RF) range. The transmitted RF signal 7060 has, for example, the following form

[0184] s(t) = u(t)cos(2πf c t + θ) (Equation 1)

[0185] In Equation 1, the carrier frequency is f c (usually in the range of 100 MHz to 100 GHz, for example 3 GHz to 12 GHz, for example 5.8 GHz or 10.5 GHz), t is time, is an arbitrary phase angle, and u(t) is the pulse shape. In a continuous wave system, the magnitude of u(t) may be single and can be omitted from Equation 1. More generally, the pulse u(t) can be defined in Equation 2 as:

[0186]

[0187] where T is the period width, T p is the pulse width. Where T p << T, which becomes a pulsed continuous wave system. In one case, when T p becomes very small, the spectrum of the transmitted signal becomes very wide and the system is called an ultra-wideband (UWB) radar or pulsed radar. Alternatively, the carrier frequency of the (chirped) RF transmitted signal 7060 can be varied to produce a so-called frequency-modulated continuous wave (FMCW) system.

[0188] The transmitted radio frequency signal 7060 can be generated by the RF transmitter 7020 using a local oscillator 7040 coupled to a circuit for applying pulse gating. In the FMCW case, a voltage-controlled oscillator is used together with a voltage-frequency converter to generate the RF signal 7060 for transmission. An antenna 7050 can be used to couple the transmitted RF signal 7060 into the air. The antenna 7050 can be omnidirectional (emitting power more or less equally in all directions) or directional (preferentially emitting power in certain directions). Using a directional antenna 7050 may be advantageous so that the transmitted and reflected energy can be sensed within the area. In one implementation of the monitoring device 7000, a single antenna 7050 with a single carrier frequency is used for both the RF transmitter 7020 and the receiver 7030. Alternatively, multiple receive and transmit antennas 7050 with multiple carrier frequencies can be used.

[0189] The monitoring device 7000 is compatible with various types of antennas 7050 (e.g., simple dipole antennas, patch antennas, and helical antennas) in various embodiments, and the choice of antenna may be influenced by factors such as desired directivity, size, shape, or cost. It should be noted that the monitoring device 7000 can operate in a manner that has been demonstrated to be safe for human use. The monitoring device 7000 has been demonstrated to have a total system transmitted average power of 1 mW (0 dBm) and less. The recommended safety level for RF exposure is 1 mW / cm 2 At a distance of 1 meter from a system transmitting at 0 dBm, the equivalent power density will be at least 100 times less than this recommended limit.

[0190] In use, the transmitted RF signal 7060 reflects from an object that reflects radio waves (e.g., the air-body interface of the patient 1000), and some of the reflected signal 7070 will be received at the receiver 7030, which may be collocated with the RF transmitter 7020, or may be separate from the RF transmitter 7020 in a so-called "dual-station" configuration. The received signal 7070 and the transmitted RF signal 7060 may be multiplied (in analog or digital form) in a mixer 7080. The mixer 7080 may be in the form of a multiplier (as shown below in Equation 3) or in a circuit that approximates the effect of a multiplier (e.g., an envelope detector circuit that adds a sine wave). For example, in the CW case, the mixed signal will be equal to

[0191] m(t)=γcos(2πf c t)cos(2πf c t+φ(t)) (Equation 3)

[0192] Where (t) is the phase term resulting from the path difference of the transmitted and received RF signals 7060 and 7070 (in the case where the reflection is dominated by a single reflecting object), and is the attenuation experienced by the reflected signal 7070. If the reflecting object is stationary, then (t) is fixed. In the monitoring device 7000, the reflecting object (e.g., the chest of the patient 1000) is typically moving, and (t) will be time-varying. As a simple example, if the chest is moving at a frequency of f due to breathing, m The mixed signal m(t) contains f m The component at (and the component at 2f ccentered component, which can be simply removed by low pass filtering). After mixing, the signal at the output of the low pass filter is called the motion signal or demodulated sensor motion signal 7003 and contains information about overall body (non-respiratory) motion and respiratory motion. An example of this approach can be considered with reference to U.S. Patent Application Publication No. US2014-0163343, published on June 12, 2014, the entire disclosure of which is incorporated herein by reference.

[0193] The amplitude of the demodulated sensor motion signal 7003 is affected by the average path distance of the reflected signal, resulting in the presence of detection nulls and peaks at the motion sensor 7010 (i.e., areas where the motion sensor 7010 is less sensitive or more sensitive). This effect can be minimized by using an orthogonal technique in which the RF transmitter 7020 simultaneously transmits a signal 90 degrees out of phase (orthogonal) with the transmitted RF signal 7060 in equation 1. This results in the generation of two reflected signals, both of which can be mixed and low pass filtered by the mixer 7080 to produce two demodulated sensor signals, which are referred to as "I signal" and "Q signal" in the corresponding I channel and Q channel. The motion signal 7003 may include one or both of these signals. Such an approach may be considered with reference to U.S. Patent Application Publication No. 2014-0163343, published on June 12, 2014.

[0194] In UWB implementations, an alternative method of acquiring the motion signal 7003 may be used. By measuring the delay between the transmitted pulse and the peak reflected signal, the path distance to the most important air-body interface can be determined. For example, if the pulse width is 1 ns and the distance from the motion sensor 7010 to the body is 0.5 meters, the delay before the peak reflection of the pulse reaches the receiver 7030 will be 1 / (3×10 8 )s=3.33ns. By transmitting a large number of pulses (eg, one 1 ns pulse every 1 μs) and assuming that the path distance varies slowly over a given period, the motion signal 7003 can be calculated as the average of the time delays over that period.

[0195] In this way, the motion sensor 7010 (e.g., a radio frequency sensor) can estimate the respiratory motion of the chest wall, or more generally, the motion of the body part of the patient 1000 that the monitoring device 7000 is monitoring.

[0196] As discussed above, the received signal 7070 may include large motion artifacts, for example as a result of overall body movement. This is because the reflected signal from the body may contain more than one reflection path and result in a complex signal (e.g. if a hand is moving towards the sensor and the chest is moving away) or if the chest is part of two separate movements - circulatory (breathing) type movements may be superimposed on the general movement of the whole body, i.e. when the user's position in bed is adjusted). Receiving such a signal is useful because it may indicate that the upper torso is in motion, which may be useful in determining sleep state.

[0197] In order to improve the quality of the respiratory motion signal and the more general body motion signal, various methods can be used to limit the body volume of reflected energy collected by the sensor unit 1200. For example, the sensor unit 1200 can be made "directionally selective" (i.e., it transmits more energy in certain directions), as can the antenna of the receiver 7030. Directional selectivity can be achieved using directional antennas 7050 or multiple RF transmitters 7020. For example, in the case of multiple transmitters, each RF transmit array element can be considered as a plurality of groups, each of which forms a directional beam and modulates a different waveform, and all beams are steerable, such as pointing in the same direction. In an alternative form of the present technology, a continuous wave, FMCW or UWB radar is used to obtain a similar signal. A technique called "time domain gating" can be used to measure only the reflected signal 7070 generated by the signal at a certain physical distance from the sensor unit 1200. Frequency domain gating (filtering) can be used to ignore the motion of the reflecting object above a certain frequency. Those skilled in the art will understand the methods of implementing these techniques.

[0198] In an implementation of the monitoring device 7000 that uses multiple frequencies (e.g., at 500 MHz and 5 GHz), the lower frequencies can be used to accurately determine larger motions without phase ambiguity, which can then be subtracted from the higher frequency sensor signal (which is more suitable for measuring smaller motions). Such an approach can be considered with reference to U.S. Patent Application Publication No. 2014-0350361, published on November 27, 2914. Using such a sensor unit 1200, the monitoring device 7000 collects information from the patient 1000 and uses this information to determine respiratory motion, as well as more general body motion information.

[0199] For each monitoring period, motion signal 7003 may be stored in memory 7002 of sensor unit 1200 and / or transmitted over a link (connection 7008) for cloud storage or for storage in an external computing device 7005. In one implementation, the duration of each monitoring period is overnight.

[0200] The processor 7006 of the sensor unit 1200 or the processor 7006 of the external computing device 7005 can process the stored one or more motion signals 7003 according to a monitoring process (such as those described in detail below). Instructions for the described process can be stored on a computer-readable storage medium (e.g., the memory 7002 of the sensor unit 1200) and interpreted and executed by a processor (e.g., the processor 7006 of the sensor unit 1200).

[0201] It should be understood that, in addition to Figure 7C-2 In addition to the single-box embodiment in FIG. 7 , the monitoring device 7000 may also be of a more distributed nature. Thus, the monitoring device 7000 may include multiple RF sensors and / or smart speakers distributed in multiple locations and linked to a central storage and / or processing unit. Such distribution of sensors will allow for effortless and unobtrusive monitoring of the user at multiple predetermined locations.

[0202] 8.3.3 Alternative monitoring equipment

[0203] As previously mentioned, still other forms of the present technology may involve an RPT device 4000 or respiratory therapy device configured to provide respiratory therapy (e.g., pressure therapy or high flow therapy) to a patient 1000 via an air circuit 4170 to a patient interface 3000, such as Figure 1 As shown. In this case, the RPT device 4000 can also be configured as a monitoring device, such as when it is configured as a processing device as described herein, when it receives sensing-related signals from a discrete device or an integrated device having a passive and / or active sensing device as described herein. Therefore, the sound-based sensing method and / or the radio frequency sensing method can be implemented in or by other types of devices such as bedside devices (e.g., respiratory therapy devices, such as continuous positive airway pressure (e.g., "CPAP") devices or high-flow therapy devices).

[0204] 8.3.4 Monitoring, screening and / or diagnostic processes

[0205] In one aspect of the present technology, the monitoring device performs a monitoring process to monitor coughing and / or sleep of the patient for screening SDB risk associated with the breathing of the patient 1000 from active and / or passive sensing signals.

[0206] The system described herein may rely on a contactless alternative to PSG or PG (polysomnography) for detecting / screening / diagnosing sleep disordered breathing (SDB) - i.e., there are no wires, electrodes, wearable devices, or mattresses required to detect SDB. This is in contrast to traditional lab / hospital-based polysomnography (PSG), which requires many sensors to be attached or worn on the subject's body. The significant problem with PSG is that it requires a lot of hardware that is uncomfortable, not readily available, expensive, and has a negative impact on your sleep (different sleeping environments, instrumented). As previously mentioned, PSG monitors many body functions during sleep, including brain activity (EEG), eye movements (EOG), muscle activity or skeletal muscle activation (EMG), and heart rhythm (ECG).

[0207] By using non-contact sensors / sensing and / or combining such devices, for example, in ubiquitous processing devices (e.g., general computing devices, smart phones, tablet computers, smart speakers, smart TVs, smart watches, etc.), a very effective screening device can be provided. In this regard, the processing device can be a device that almost everyone already owns, which allows for a relatively inexpensive, widely available, portable, and computationally fast solution. Using such a system, SDB screening results can be almost instantaneous (i.e., after the sleep period is completed), and in some versions, an all-in-one solution can be provided that does not require specialized personnel or separate tools for analysis. Such a solution can provide longitudinal monitoring and can allow for the integration of additional metrics and risk factors.

[0208] An overview of some of the advantages of some versions of this technology:

[0209] In some versions of the present technology, the screening system may advantageously combine processes for SDB risk assessment and sleep segmentation assessment to improve the accuracy of either assessment. For example, when estimating apnea / hypopnea counts (AHI), actual sleep parameters may be used to improve accuracy (e.g., allowing apneas to be detected while a person is asleep (in sleep stages), and / or adjusting for sleep types associated with people with OSA). This may alleviate some issues with previous systems, in which micro-arousals may cause the system to indicate an awakening interval, which may result in the possibility of missing SDB-related events. For example, the event may be ignored by the system because the system determines that the person is awake, rather than in sleep with micro-arousals.

[0210] The system, for example, when considering the intensity of SDB as described herein, can also consider obstructive and central apneas in its assessment. In systems that only provide snoring detection, such apnea or hypopnea events are ignored. In contrast, the method implemented by the described technology can detect both because the system can measure / sense chest displacement (e.g., movement in the absence of contact sensing) to indicate actual breathing effort. Therefore, if the monitored subject stops breathing (i.e., central apnea-where the brain (respiratory control) does not send a signal to breathe), this absence of breathing movement can be detected. Conversely, when there is an obstructive event (e.g., obstructive hypopnea or obstructive apnea-where the monitored subject is making (movement) efforts to breathe against a partially or completely closed airway), the system can still "see" (detect or sense) modulation. In passive audio analysis, efforts can be detected by sensing that a noise period follows the cessation of noise. This can also be considered in conjunction with active sensing of breathing effort (movement) during such presence and absence of noise.

[0211] Thus, as described herein, the apparatus and methods may apply active sensing, such as using any radio frequency sensing system or radar, such as by using pulsed continuous wave (CW), continuous wave (CW), frequency modulated continuous wave (FMCW), ultra-wideband (UWB) sensing methods - where the system can detect modulations in breathing effort (user motion). Similar motion sensing may be achieved with active sound sensing systems (sonar). If a microphone is also available, the sensing technology may be achieved by merging passive acoustic sensing technology with at least one active sensing technology.

[0212] Multi-sensor applicability:

[0213] For example, the SDB screening system of the present technology can implement a variety of non-contact sensing methods - including active acoustics (e.g., low frequency ultrasound - sonar - whereby acoustic reflections are processed), passive acoustics (e.g., using one or more microphones to "listen" for audible breathing sounds), and / or electromagnetic (e.g., radio frequency (RF) or radar - where processing is performed via RF reflections). Thus, in some versions, the acoustic version can be implemented with a combination of both active acoustic sensing (using speakers and microphones to process reflections) with correlated signal analysis and passive acoustic sensing (listening) with correlated signal analysis. In some versions, active sensing can be implemented with an RF system, (a) without passive acoustic analysis, such as when a microphone is not available, or (b) using passive acoustic analysis in a fusion approach that implements RF sensing and microphone sensing. When active sensing is employed, sensing can be implemented at multiple ranges (e.g., different motion sensing channels at different distances from the sensor), such as by implementing pulsed CW, FMCW, and UWB.

[0214] Cross-scope processing:

[0215] As discussed herein, the system can be implemented to evaluate motion signals across multiple sensing ranges (so that it can combine motion information in signals from multiple locations (distances) in the sensing area or sensing vicinity of space). This combination can involve sensing technologies such as FMCW (Frequency Modulated Continuous Wave - Sonar or RF) or other ranging technologies to produce different motion channels / range bins for different detection distances. By evaluating motion information across ranges with such different range channels, a multi-range approach can be used to generate an effort signal.

[0216] In the example of active processing, phase differences associated with motion in different sensing ranges can be ignored, for example by taking the absolute values ​​of different slices in the space of the active 2D signal volume. Consider that a detection signal (channel) or range bin from a "30 cm" range may have similar motion information, but may be out of phase with such a signal at a "35 cm" range. By combining such channels (e.g., taking the absolute values ​​of different range bins before combining), it is possible to address the challenge that such signals may be out of phase. In summary, this is a potential reason for using a corrected weighted sum over range bins to generate an active respiratory effort signal.

[0217] Thus, the system can be configured to calculate the envelope of the respiratory effort signal across multiple active channels (e.g., where each such active channel is derived from an FMCW signal corresponding to a space). Similarly, the envelope of the effort signal can be derived using multiple channels derived from an acoustic passive sound analysis, where such channels derived from such passive sound analysis can correspond to different audio frequency bands. These different frequency band channels are not technically strictly related to the range of distance. However, the range / distance can be associated with such frequency bands, such that the amplitude in different frequency bands may change due to distance.

[0218] Identify SDB event clusters (a set of tightly grouped events):

[0219] As discussed herein, the system can also be implemented to identify clusters of apnea events rather than individual occurrences. For example, by using an in-band frequency metric (IBM), the signal modulation strength can provide a signal quality (modulation strength) that can be assessed to identify clusters of these events. Thus, the frequency of such clusters can provide a basis for calculating or assessing SDB risk, and can be performed through active sensing techniques such as using sonar acoustic sensing.

[0220] When considering the effect of many apneas, the intensity of the modulation may be considered in relation to the instability of the detected signal over a longer period of time. For example, such modulation may be considered in relation to a period of time where many apneas may occur, compared to considering only the period of time of the signal as one apnea. The intensity of the modulation of the sensing signal (which may be comparable to the amplitude), where the modulation intensity is attributable to many apneas, such as due to severe apneas, provides an assessment feature that considers many apneas simultaneously. It may provide insight into the stability (or instability) of the body's physiological control system.

[0221] Another implementation could simply detect each apnea event, classify the type of apnea, and then count them on an individualized basis. This approach would not be equivalent to the "modulation intensity" approach.

[0222] It has also been found that the modulation frequency of the determined respiratory effort envelope has subject-specific characteristics that can be predicted when estimating the SDB level of a given subject. Therefore, this metric can be implemented as a feature for SDB-related risk classification by a classifier.

[0223] With respect to passive sensing and analysis (e.g., where the system uses a microphone to record a person's breathing-related sounds), a passive envelope generation process can also be implemented to evaluate the modulation intensity. This approach is not common in the art because it does not process snoring noise in a typical or standard manner. When considering the modulation intensity of the noise (whether it is breathing or snoring), the features of interest can be based on consideration of baseline noise variations across frequencies (e.g., 9 audio bands can be generated). Through signal processing, the frequency bands can be processed to remove baseline trends (remove artifact trends). The resulting detrended signals can be averaged. Therefore, changes in intensity can be evaluated. The method (e.g., including trend reduction and averaging) is robust so that it can avoid the effects of shorter time scale background noise that may be simultaneously recorded by the microphone.

[0224] An envelope normalization step (which can be a combined mixer approach) can be used for both active and passive processing. Because each signal is "pulsatile" in nature (appears in bursts), the system can first integrate the signal and then high pass filter the result. This "pulsatile" or pulsating signal type can benefit from pre-processing applied to the passive audio breathing effort signal. For example, using passive audio, a system that measures the sound of breathing / snoring can include such a signal that is pulsatile in nature when the system "hears" the exhalation or inspiration portion of the user's breath.

[0225] Subsequently, a normalization step may then divide the resulting signal by the running standard deviation mixed in (i.e., added to a version of itself (which may be analogous to a "fade knob" in sound processing). The motivation for envelope normalization is to reduce the amplitude and therefore the effects of large motion artifacts that may occur in the time window of the signal. This may make the frequency domain visually "cleaner" for subsequent processing, so that the target sleep disorder breathing event information may be more easily evaluated by the system.

[0226] A pre-processing step of integration and high pass filtering (which may also be described as band pass filtering) provides for transforming the pulsation signal into a shape more suitable for spectral analysis (e.g. making it more sinusoidal in nature). This processing step may be, but is not required to be applied to the sonar effort signal, as the measured chest displacement does not typically exhibit the same pulsation characteristics.

[0227] This can be considered as a variable normalization method, which can result in better accuracy for the spectrum analysis processing method.

[0228] The system process can then apply the clusters identified by sonar (and / or RF) to ensure that the sleep staging output (e.g., hypnogram of different sleep stage times) is correctly scored for sleep (not wake), such as for situations where SDB is identified or measured, particularly in cases of severe SDB. Thus, the spectral approach does not need to identify or detect individual events, such as on an event-by-event basis. Instead, it can use a classifier based on overnight trends to characterize modulations associated with SDB that are typically from clusters of events.

[0229] Modulation of breathing effort:

[0230] As described in more detail herein, the system can detect respiratory effort modulation to assess or detect the presence of SDB, and include obstructive events without central events. The system is also capable of detecting periodic motion associated with SDB recovery / wake-up. This is useful when the respiratory signal and / or effort modulation is weak. For central events, the modulation is close to 0 and therefore cannot be included / reflected in the modulation intensity signal described herein. For hypopnea, this is similar to AM (amplitude) modulation. In obstructive events, more severe AM modulation is detected.

[0231] In summary, the system can detect all forms of SDB events. As described in more detail herein, the system can detect modulation of respiratory effort in order to assess or detect the presence of SDB events, including hypopnea, obstructive, central, and mixed events. All SDB events will ultimately appear as AM modulation in the respiratory effort signal envelope. The AM modulation intensity of an obstructive event is typically greater than the AM modulation intensity of a hypopnea. The system also enables detection of periodic motion associated with SDB recovery / wake-up, which may be useful when the respiratory signal and / or effort modulation is weak. Therefore, the system's evaluation of AM modulation (e.g., using one or more thresholds) can allow it to distinguish or detect such events.

[0232] For example, the system is able to detect obstructive, hypopnea, and central events through the same processing of the respiratory effort signal. All of these events can introduce modulation in the effort signal, although their nature may be different. Obstructive events typically cause an increase in effort (i.e., breathing against a restricted airway). Hypopnea is similar to obstructive events, although the airway is only partially blocked and the change in effort is smaller. In contrast, central events will appear as a decrease in effort relative to the baseline (i.e., no attempt to breathe). Event types are verified based on the "gold standard" scoring performed in the PSG laboratory. Therefore, thresholds for evaluating AM modulation can be derived through system training. In addition, event types can be classified by evaluating whether there is an associated increase or decrease in effort (also a change in modulation shape and amplitude) relative to the recent historical baseline.

[0233] Steps to achieve this on an active radar or sonar (e.g., in the case of a demodulated 2D matrix representing range bins of an FMCW front-end system as described in International Patent Publication No. WO2018050913) include a process of combining signals from different range bins (within a defined sensing range), such as by taking a corrected weighted sum. As described in more detail herein, with such a signal combination, the system can then, for example, search or determine the intensity of the modulation, such as by using a moving window of the spectrum, in order to calculate a spectral quality metric.

[0234] For passive sensing stream processing, e.g., where a signal received by a microphone (i.e., sound) is analyzed, different pre-processing steps may be performed to estimate the envelope associated with active sensing signal processing. In this regard, different transforms may be applied because the signals differ in composition. Thereafter, the process may integrate the signal (e.g., envelope from an active or passive processing stream) and then filter, e.g., by high pass filtering.

[0235] The basic principle of this process of envelope detection of passive signals can be considered in conjunction with the following description. Since there are typically gaps (e.g., reduced sounds) associated with sound between breaths, the system can be configured to identify peaks approximately 5 seconds apart, and interpolate signals between these peaks to create envelopes. Subsequently, the envelope can be integrated and high-pass filtered to process the "pulsating" characteristics of the signal. This can be similar to the peak and hold method, but can provide more suitable performance. The peak search and interpolation process can achieve initial cleaning of the signal, but they may still be very jagged in appearance (e.g., related pulses at the end of the event - related to recovery breathing). Therefore, additional integration and high-pass filtering steps can be applied to resolve such residual sawtooth. Another method can be to apply a bandpass filter. However, integration processing is useful because it can transform the pulse type morphology so that the resulting signal is closer to a sine in appearance or composition.

[0236] The actively and passively derived breathing envelopes can then be normalized in a similar manner. This can help prepare the signal for high-precision spectral analysis, which can provide a clear point of differentiation in terms of performance. This can be an important step, for example when the sonar envelope scale varies greatly throughout the night (e.g., motion can cause very large spikes / artifacts), and this normalization can help compensate for such artifacts.

[0237] Such an envelope may also provide the frequency, e.g., median frequency, of respiratory effort modulation relative to the active respiratory envelope. (e.g., a power spectral density (PSD) operation may produce spectral peaks with sufficient signal-to-noise ratio (SNR) compared to the noise floor) Thus, such information may be evaluated as input features for AHI and / or risk probability classifiers. For example, the method allows the system to be configured to extract modulation frequencies over nearly any time period (e.g., typically greater than a single event). This also applies to passive envelopes. Optionally, the median of the active / sonar modulation frequencies for the entire night / recording may be used for noise reduction.

[0238] Combined active and passive features:

[0239] The advantage of processing active and passive sensing signals is that they both contain different information about a common sensing target. For example, they measure different aspects of the physiological signal that are the basis for estimating the presence of different types of SDB. For passive sensing signals, the sounds of snoring and / or breathing are analyzed, including the intensity over different frequency bands. For active sensing signals, the displacement of the chest is measured (i.e., the system can sense chest movement when there is good signal quality).

[0240] In some cases, active sonar analysis may have difficulty (estimated signal quality is lower) when the subject's back is turned towards the sensor, and the person is under some quilting (blanket / quilt). For this situation, passive analysis can still sense snoring and / or breathing sounds (including erratic breathing, and air turbulence caused by breathing). The system is also able to detect periodic motion associated with SDB recovery / awakening. Here, passive analysis may be useful when the actively detected breathing signal and / or effort modulation is weak.

[0241] The combination of channels may be optimized for the features derived from each method based on the determined signal quality (thus emphasizing channels with better signal-to-noise ratio (SNR) or better quality).

[0242] For situations where both are deemed of acceptable quality, the system can check how well the two channels are correlated and use that information to estimate centrality or obstruction, as well as the location of the person.

[0243] The system may detect snoring by proxy, for example, when audio intensity modulation in an audio band (eg, between 250 Hz and 8 kHz) reflects the presence of snoring and use such relative acoustic intensity to determine the SDB indication described herein.

[0244] The system can detect a person's breathing sounds and, optionally, evaluate breathing sounds or snoring as a basis to rule out the occurrence of central apnea, as they are not typically associated with snoring, such as when breathing sounds are detected at approximately 5kHz. Such frequency components associated with snoring tend to disappear during central apnea events.

[0245] Slow periodic modulation due to environmental interference can cause problems with passive / active sensing. Because the system uses two different sensing modalities, there is less chance that both will have poor or no physiological signal information at the same time. This is of course different than a situation where the user leaves the room. The system can sense this happening and trigger an "absent" mode in this situation, which can pause SDB screening until the user is "present" again, and resume SDB monitoring when the user returns and is asleep.

[0246] Some advantages over oximetry and capnography:

[0247] Examples of the system described herein can estimate wakefulness time based on small movements at the end of the event, which is an advantage over other sensing such as SpO2 that may not do so. The advantage over oximetry and capnography is that the system is able to measure partial or full movement of the body that may occur as part of SDB-related wakefulness through sonar or audio. Other sensing modalities may only detect movement localized to the sensor location, which may not be as useful for SDB detection.

[0248] The system not only measures chest modulation, it also detects motion due to recovery events, which is not usually visible during capnography.

[0249] In addition, the system can achieve the distinction between sleep / wake, for example, using SDB-related corrections. This helps SDB detection accuracy because AHI is defined as sleep events / hour.

[0250] How to avoid detection of a second user in bed:

[0251] For the case of two people in the bed, the system can consider multiple features to separate the closest person (the person to be analyzed) from the farther bed partner. For the closer person, the detection signal can be stronger. As described in more detail herein, the system can implement a weighted average of the signal. Such processing tends to favor stronger signals (by emphasizing it) and tends to suppress weaker signals. Therefore, the processing can separate information from closer people from information from farther people. In some cases, if a person snoring is not the person closest to the sensor, snoring can be sensed auditorily, but the associated movement may not be sensed. This inconsistency can be used to separate (and optionally reject) partner snoring. In other words, passive sensing analysis may be inconsistent with active analysis, which can serve as a basis for denying the SDB conclusion. For example, a voting mechanism such as implemented with controller / processor logic can be used to select a signal with a better signal-to-noise ratio over a period of time. Because active analysis can contain distance information (e.g., FMCW-based processing), it can be an effective means of rejecting data from a farther partner who is snoring. For example, a weighted sum of quality-based estimates can be used. If the two methods directly contradict each other over multiple time scales (which is very unlikely), the system can flag an error condition.

[0252] For active sensing that produces varying range detection (range bins), different range bins can also be used to localize the sensing area to the nearest person. This can even help if the person rolls to the far side of the bed (e.g., away from or close to the sensor).

[0253] The system can also implement natural shielding, because in the case of a phone or sensor placed on a nightstand / bedside table, a closer person "shields" (partially blocks) a farther person; this applies to active sensing. This shielding can be facilitated by positioning and orienting the sensor toward, for example, the chest of the first person, making it difficult for another person to move into the sensing area. When the smart device is placed on a bedside locker or table, etc. near the first person being monitored, shielding is a natural result of the sensing modality. That is, the primary or nearest user will partially or completely physically block the active sensing signal (e.g., sonar waveform) from "impacting" / illuminating the secondary user, so their movement / breathing is not measured, or is weak. This is also true for passive audio to a lesser extent, because the secondary user is further away, so their audio signal will generally be weaker. If the monitoring device is placed behind the bed, such as on a shelf, facing down toward the user and bed partner, shielding can be reduced (increasing the chance of the partner being detected). The FMCW active system as described herein can be configured to distinguish two people by segmenting multiple distance bins.

[0254] The system can also take advantage of the fact that many modern smartphones have multiple speakers and microphones (e.g., a bottom-side speaker box), as well as an earpiece receiver that can also act as a second speaker for stereo playback and also has a bottom-side microphone, and one or more microphones located on the top edge or top back (portable camera) of the phone.

[0255] For example, a user may want to listen to sleep sounds (e.g., ocean waves synchronized with their breathing) to help them fall asleep in order to relax. A common problem with smartphone speakers is that by mixing the sonar sensing sound with music or other audio sources, the transmit power of the sensing signal is reduced, and very small speakers and associated smart power amplifiers may produce unwanted audible artifacts that are not in any source signal due to non-idealities in the speaker and / or amplifier chain (even if the music has been low-pass filtered to remove any frequencies that may overlap with the sensing waveform). One solution may be to separate the channels for separate speakers. For example, the system may route only the filtered music to the second speaker (which may require mixing a stereo music source to form a single channel representation), and play the sonar sensing sound through the first speaker - thereby maintaining sensing fidelity as well as music / audio playback fidelity. Another approach is to use two speakers for sensing, but each speaker has a different sensing frequency range and / or repetition rate. Reflections may be monitored by one or more microphones. This may be beneficial when the user uses more or more thick quilts (folding quilts) and / or blankets and the signal level is low, or for ceilings or wall materials of different heights. Typically, two or more smart device speakers are pointed in different directions, and different multipath reflections can be exploited to better detect the user's chest and body movement. The reason for using different frequency ranges (either in absolute terms, or in time) is to allow the reflections to be processed directly (potentially without the need for an encoding scheme); in addition, because speakers, microphones, and associated amplifiers are typically non-linear in the sensing frequency range above 18kHz, the second reflected echo signal provides an additional source of information from which to derive breathing and movement. It can also be seen that if a second speaker is used to play sound while the user is asleep, the music can be faded out and replaced with the secondary sensing signal once they are sound asleep.

[0256] Sleep segmentation ensures sleep-related breathing disturbances

[0257] As part of the AHI calculation, the system takes into account sleep detection so that only reduced or stopped breathing is processed during sleep. Therefore, by including a sleep segmentation process, such as described herein and in International Patent Publication No. WO2015006364 (PCT / US2014 / 045814), the accuracy of the system is improved.

[0258] Furthermore, the number of apnea events may be defined relative to total detected sleep time, rather than a simpler assessment that considers the number of apnea events within a detected period of sleep that may not be limited.

[0259] PLM / RLS:

[0260] Periodic limb movements (PLM) and restless legs syndrome (RLS) may appear in the modulations detected by the system. However, these modulations are not as consistent as apnea clusters and can occur on different time scales. Therefore, the system is robust to moderate levels of PLM, whereby the repetition / occurrence rate of PLM is higher than the repetition / occurrence rate of apnea and therefore appears different. In addition, PLM and RLS are unlikely to interfere because, depending on the angle and position of the phone or sensor, a person's limbs (including legs) are typically outside the main sensing area. Optionally, a separate or combined RLS / PLM detector may also be used with the system of the present invention to track these periodic movements. Examples of such detectors may include any of the methods described in PCT / EP2016 / 080267, the entire disclosure of which is incorporated herein by reference.

[0261] Signal combination using passive and active sensing signals can help mitigate the effects of PLM. For example, the system can implement greater weighting for distance bins associated with motion where breathing occurs (e.g., weighting can be based on breathing intensity). Thus, system processing can help prevent PLM from becoming a significant confounding factor for the system.

[0262] Teeth Grinding / Bruxism:

[0263] Tooth grinding / bruxism occurs at higher (faster) rates that do not fall within the modulation range being addressed. Therefore, such events can be avoided in the SDB detection method described herein.

[0264] Central and obstructive apnea:

[0265] The system can distinguish between central and obstructive clusters based on the shape of the parameters, and based on the characteristic relationship between the passive sensing signal characteristics and the active sensing signal characteristics. Central and obstructive events can be detected as different types of modulation in the respiratory effort signal. Obstructive events typically cause an increase in effort (i.e., breathing against a restricted airway). Obstructive hypopnea is similar to an obstructive apnea event, but because the airway is only partially blocked, the change in effort is smaller. In contrast, central events will appear as a decrease in effort relative to a baseline (i.e., no attempt to breathe). Therefore, the system can detect events and classify events by type by evaluating whether they have an associated increase or decrease in effort, a characteristic shape or contour of the effort envelope, and / or a relative change in amplitude relative to the baseline effort (local or global time change).

[0266] In some cases, the device can be implemented to use snoring or other characteristic audio features (e.g., gasping, choking) to determine that the event is obstructive in nature. For example, the system described herein can implement snoring or other sleep disordered breathing event monitoring as described in U.S. Patent Application Publication No. US-2018-0256069-A1 published on September 13, 2018, the entire contents of which are incorporated herein by reference. Optionally, based on the number of central versus obstructive versus mixed events (and taking into account apneas or hypopneas), the system can be configured to recommend different types of treatment.

[0267] Position changes can be detected:

[0268] The system can detect the relative position of the nearest person in the bed, for example, if they are facing the sensor, facing away from the sensor, or on their back or stomach. Examples of sensor configurations and methods such as detecting position can be considered with respect to International Patent Publication No. WO / 2017 / 032873 published on March 2, 2017, the entire disclosure of which is incorporated herein by reference.

[0269] Inhale and exhale:

[0270] In some versions, the system may implement RF detection (e.g., Doppler sensing) to separate or differentiate between inhalation / exhalation efforts (inhalation is always toward the sensor when a person inflates). This distinction may be achieved, for example, by processing the Doppler shift (e.g., detecting changes in direction of motion), thereby yielding a direction of travel of the target, which may be mapped to inhalation / exhalation. A method for detecting changes in direction of motion, which may be implemented to differentiate between these breathing phases, is described in International Patent Publication No. WO 2016 / 170011, published on October 27, 2016, the entire disclosure of which is incorporated herein by reference.

[0271] Generate AHI:

[0272] The system can provide an estimate of the apnea-hypopnea index (AHI) or respiratory disturbance index (RDI). For example, this can involve the number of apnea and hypopnea events or the number of apneas and events per hour of sleep. Therefore, it can be seen that if the system can estimate different types of apneas (e.g., direct counts of apnea and hypopnea events) and sleep time, the AHI value can be estimated. Another approach is to provide larger-scale modulation data (e.g., training or sequences representing apneas and / or hypopneas) to a machine learning model based on estimates provided by one or more skilled human PSG scorers (labeled data), and develop a classifier (e.g., a logistic regression model classifier) ​​to estimate the AHI from these features. The system can provide estimates of the AHI and / or RDI and / or a metric that quantifies the risk of SDB (e.g., a probability output of a logistic regression model).

[0273] In combination with other risk factors:

[0274] The output of the system can be combined with other risk factors - age, BMI, race, income, comorbidities, alcohol consumption - to calculate an overall "risky sleep" index. This can be input to and characterized by the classifier described herein.

[0275] 8.3.4.1 System Processing Architecture

[0276] refer to Figure 8 An exemplary system architecture having the modules described herein may be considered, for example. Such a module may be implemented by any processing device described in this specification, such as an RF sensor-based system and / or a mobile phone. Alternatively, the processing module may be implemented separately from the sensor by a processing device located nearby (e.g., a mobile phone or tablet computer located near the RF sensor) or at a remote server. Typically, the system performs an assessment of SDB risk, which is derived from sensing signals associated with one or more sensing periods (e.g., one night of sleep or multiple nights of sleep) of the sleeper, such as from acoustic and / or radio frequency-based sensing devices described herein. One or more processes may be configured to generate an SDB risk indication to provide any one or more of the following:

[0277] (1) A binary classification flag (true or false) for exceeding a clinical threshold for SDB events (e.g., AHI greater than a threshold such as 15). For example, the threshold can be adjusted to 10 or 5, or higher or lower as desired. Optionally, the binary classification flag can be symbolic. For example, such an indication can be a color-coded output to indicate one color for true and another color for false (e.g., "green" for counts below a threshold (e.g., 15), and red for counts greater than or equal to the threshold).

[0278] (2) a calculated probability of SDB risk (e.g., as a percentage or scaled indication within a range such as 1 to 100, e.g., where one end of the range indicates less SDB and the other end of the range indicates more SDB);

[0279] (3) An estimated score of relevant SDB events (eg, apnea-hypopnea index (AHI) or other count of SDB events).

[0280] In addition, the system can be configured to detect sleep stages and / or provide a sleep score, a mental renewal score / indication, and / or a body renewal score / indication, as described herein, and generate an output indicating a sleep stage based on motion detection. For example, this can be implemented as described in International Patent Publication No. WO2015006364 filed on July 8, 2014, the entire disclosure of which is incorporated herein by reference. In addition, the system can be configured to implement corrections to the sleep stage analysis to ensure that such segmentation determination is correct or more accurate - even for people / users with severe SDB. In an example described in more detail herein, this can be implemented using a method (machine algorithm) to implement a sleep / wake correction mask, such as a time series representing an indication of sleep or wake determined from an analysis of a sensed signal. This correction can be based on detecting and evaluating the intensity of modulation detected within the SDB frequency range (e.g., modulation occurring within an interval of approximately 25 to 100 seconds).

[0281] In this regard, events tend to occur in succession, and the time period between two consecutive events is typically (based on empirical analysis) in the range of 25 seconds to 100 seconds. The system can, for example, be tuned to detect periodicity (i.e., modulation) within a specific range. When the system detects a "cluster" of SDB events, sleep / wake correction is applied. The events in a cluster occur at a certain rate (or frequency). The system can also be configured to ignore events whose periodicity falls outside the typical "normal" physiological range associated with apnea based on the fact that these events are more likely to be spurious. Separate RLS and / or PLM detectors can be used.

[0282] like Figure 8 As shown, for example, the processing using one or more algorithms can be implemented as multiple modules (for example, three modules in this example). The operations / functions of the modules implemented by one or more processors are described in detail in the following sections and can be used as reference for Figure 7B-288 is implemented as part of the method shown. In this example, the operation modules can be implemented as (a) an extraction module 8810, which can be an online (where "online" in this case can be understood to mean using a processor to process data blocks or frames as they are collected, so as to give the fastest response to the user at the end of the night and avoid potentially buffering many gigabytes of raw audio data) module, (b) an evaluation module 8820, which can be an offline (where "offline" in this case can be understood to mean "end of period" processing of intermediate parameters provided by previous "online" processing steps) module, and (c) an output module 8830, which can also be an offline module for output generation. Each module is described in more detail herein.

[0283] Typically, the extraction module 8810 can be configured to perform signal extraction, such as extracting signals from active sensing signals (e.g., RF and / or sonar) sensing signals and passive sensing signals (e.g., ambient audio / sound). Such a module can include, for example, multiple channels, such as for extracting I&Q (in-phase quadrature) breathing effort signals from input sensing signals (e.g., raw motion signals from RF and / or sonar motion sensing). The module can also have multiple acoustic channels, such as from passive detection, to isolate the audio energy of interest, thereby extracting it into one or more frequency bands of interest (e.g., 9 channel bands).

[0284] Typically, the evaluation module 8820, using the input signal from the extraction module, can determine modulation characteristics, such as the frequency and intensity of the modulation. The module can also calculate a sleep / wake correction mask based on the determined modulation intensity. Such a correction mask can be input to a sleep segmentation process that determines the sleep stage (e.g., a hypnogram signal) so that the correction mask can be applied to correct or avoid incorrect wake-up indications due to SDB. The evaluation module 8820, for example, with intensity, frequency, and sleep segmentation information (e.g., a hypnogram signal corrected by mask conversion), can be applied to SDB summary assessment. In such an assessment, SDB features (e.g., scalar values) can be derived from respiratory effort signals such as active sensing (e.g., sonar and radio frequency sensing) and frequency bands from passive acoustic sensing. These features can be applied to one or more classifiers. For example, a logical classifier can classify scalar values ​​to produce binary classification flags and risk probability indications as previously discussed. As a further example, a linear regression classifier can classify scalar values ​​to calculate a respiratory event score (e.g., an apnea / hypopnea index that counts these events). This calculation of a respiratory event score may be applied as an adjustment factor, such as an SDB penalty factor, which may adjust one or more portions of a sleep score, such as for use as an adjustment factor in a body refresh score / indicator and / or a mind refresh score / indicator.

[0285] Despite Figure 8 The operations are shown in the example of FIG. 1 so that the operations are divided into real-time or near real-time (during the recording of the sleep period) and at or after the end of the recording and / or sleep period, but other separations of operations may be implemented. The division of operations in the illustrated example may make particularly efficient use of the processor, such as when implemented by a smartphone.

[0286] Nevertheless, online and offline operations can be implemented so that they are all treated as offline processes, or aspects of offline operations can be implemented as online processes. As shown, the separation of online and offline is designed to effectively utilize multi-core smartphone processors, but can be adjusted based on the characteristics of the available processors and the availability of storage / memory.

[0287] 8.3.4.1.1 Extraction module 8810

[0288] The purpose of the extraction module 8810 (e.g., by implementing continuous or periodic processing, whether during sleep or after) is to regularly calculate relevant information signals based on active sensing signals and / or regularly calculate information signals based on passive sensing signals (e.g., non-contact sensing). This extraction processing will depend on the nature of the sensing signals, for example, whether they are radar (i.e., radio frequency) generated and / or sonar generated.

[0289] For example, for active sensing information signals, the extraction process can generate a breathing effort signal based on non-contact respiratory motion sensing. Such a breathing effort signal can be generated in the case of sonar as previously described herein, and / or according to the method described in International PCT Patent Publication No. WO2018 / 050913; and, in the case of radar as described herein, and / or according to the method described in International PCT Patent Publication No. WO2014 / 015238 published on October 22, or U.S. Patent No. 8,562,526.

[0290] In one example, one or more breathing effort signals can be generated from a sonar or RF demodulated signal by, for example, generating two channels (I and Q) of 16 Hz breathing effort signals. Optionally, the system can implement other sampling rates. For the case of frequency modulated continuous wave (FMCW) active front-end sensing (audio or RF), the extraction module is tasked with performing FMCW breathing effort detection. This active sensing signal extraction processing can be implemented using one or more active sensing signal extraction submodules 8812. This submodule can be implemented using a sonar sensing signal extraction submodule and / or a radar sensing signal extraction submodule.

[0291] Additionally, for passive sensing information signals, the extraction process of the extraction module 8810 can extract information signals from the acoustically sensed audio signal. Such passive audio energy can be extracted from a full-band audio signal (e.g., the detectable audio range of the implemented microphone) and put into SDB related frequency bands (e.g., 9 channels with different frequency bands). Such signals can be generated at 1 Hz, but other sampling frequencies can be implemented as needed. Such passive audio signal extraction processing can be performed by the passive sensing signal extraction submodule 8814.

[0292] In some cases, the extraction module 8810 may further implement modules or submodules for extracting a respiration rate signal and / or a signal quality signal. The respiration rate signal represents the respiration rate over time, and the signal quality signal represents the quality of the recorded sensed signal over time.

[0293] 8.3.4.1.1.1 Extracting the respiratory effort signal from the active sensing signal

[0294] You can refer to Fig. 9 , 10 and 11, consider active sensing signal processing, for example, for generating / extracting a respiratory effort signal (based on respiratory motion) in the 8810A. In some versions, such sensing signal processing may include processing that combines complex demodulated sensing signals from multiple ranges (range bins) (see Fig. 9 ), where each range bin signal can be divided into two channels, one for the real part (which can be the I signal) and the other for the imaginary part (which can be the Q signal). Usually, there are two inputs that meet the needs of extracting the respiratory effort signal:

[0295] a. The sonar signal in each range bin over a period of, for example, 5 seconds

[0296] b. Respiration analysis for each of these range bins.

[0297] The breathing analysis is based on a measure of the highest spectral power density in the breathing frequency range, telling us how much breathing signal is present in each range bin. This breathing analysis can be performed at each time interval (e.g., 5 seconds) (spanning signals over longer periods, such as 64 seconds). For each range bin, we will get a measure output at the left and right boundaries of, for example, the 5-second time domain.

[0298] A way of combining all range bins into a single channel via a respiration-based metric may include a method having the following steps:

[0299] a. At the left and right boundaries of the time domain, the metric value corresponding to a specific distance bin is divided by the sum over all distance bins to define the weight of each distance bin;

[0300] b. For each distance bin, linearly interpolate the weights between the left and right

[0301] c. Multiply the weights by the signal in each range bin

[0302] d. Perform the above summation for all distance bins

[0303] This analysis can be performed in time with the sonar respiration analysis, for example, both performed over a 5 second period, and using the signal from the range bins including the respiration information to extract the effort signal. Fig. 9 The exemplary sensing signals include signals from distance bins at 0.3 m to 3 m.

[0304] This treatment may help resolve any of the following issues:

[0305] (a) Between range bins, the signal may undergo a 180° phase change;

[0306] (b) range bins that do not contain modulations within the respiratory range (i.e., respiratory movement frequency) must be excluded;

[0307] (c) Continuity between consecutive 5-second segments should be ensured to account for possible changes in the subject's posture.

[0308] Therefore, if Fig.10 As shown, such processing may include applying absolute value processing to each discrete range signal or range bin and then combining the range bin signals based on a weighted average of the respiration-related frequencies derived from the PSD operation. This process may be performed separately for each of the I and Q channels. Consider an example, subject to the following processing:

[0309] (1) For each range bin, at the left and right boundaries of the 5-second time domain, the weight can be calculated based on the metric value that represents the highest power spectral density (PSD) in the breathing range;

[0310] (2) Linearly interpolate between the right and left boundaries to spread the weights across the time domain;

[0311] (3) Distance bins can be combined in the following ways:

[0312] (a) Take the absolute value of each distance bin;

[0313] (b) Take the weighted average of the above absolute values ​​by multiplying with the weighted array.

[0314] For this example, the weights are simply scaling factors applied to each range bin signal. This "weights" certain range bins proportionally compared to other range bins. When evaluating respiration rate, normalization may be required to suppress large amplitude modulation components. However, the amplitude modulation (AM) component is retained for SDB analysis because it includes relevant modulation related to SDB on a longer time scale.

[0315] 8.3.4.1.1.2SDB Passive Audio Signal Extraction

[0316] You can refer to Fig.12 The spectrum of the audio signal shown in Figure 8 is used to consider, for example, passive sensing signal processing for generating / extracting frequency bands in submodule 8812. This submodule transforms the original full-band audio signal to a frequency domain, for example, for later generating energy measurements of different frequency bands. SDB passive audio signal extraction can involve separating the audio signal into frequency band signals (e.g., 9 discrete frequency bands, dividing a range such as about 100 Hz to 8 kHz or 250 Hz to 8 kHz). In some versions, eleven frequency bands can be implemented, for example, where the frequency bands are nominally spaced at intervals of 250 Hz within the range. The extraction process can be performed once per second (1 Hz). This can be achieved by a Fourier transform (e.g., a discrete or fast Fourier transform).

[0317] For example, extraction may include the following processing:

[0318] (a) An FFT (e.g., "4096 points") may be calculated on samples of audio data (e.g., 3000 samples), e.g., 16 times per second. Optionally, this may be performed after "de-sense" and spectral windowing have been applied. In some cases, median removal may be used (e.g., performed on a per-second basis) rather than mean removal. Other sizes may be used, e.g., a 1024 pt FFT may be calculated on 1000 samples of audio data, etc. (e.g., non-overlapping frames).

[0319] (b) The energy in each frequency band can be determined by calculating the sum of the absolute values ​​of the corresponding FFT coefficients for a particular frequency band. Each such sum can be considered as an energy value for each frequency band. For the example case of 9 frequency bands, each FFT operation will result in the calculation of 9 energy values.

[0320] To extract each frequency band signal for SDB evaluation, an average value can then be determined (calculated) based on the energy value of each frequency band during a window period of, for example, one (1) second. In the example of 16 FFT operations per second, 16 energy values ​​will be averaged for each frequency band. Therefore, when the process is repeated, a separate signal is generated over time for each frequency band (e.g., 9 frequency bands for SDB and respiration), where each sample of the signal of a given frequency band represents the average energy from the time interval. The system can remove the linear components of the 9 frequency bands and then take the average. Then, envelope peak finding is performed, peak insertion is performed, etc. (e.g., by integral generation and high pass filtering), which will be described separately. For respiration, each frequency band is processed independently for respiration through preprocessing and autocorrelation. When there is a peak highlight, the autocorrelation output is used together with the signal quality metric to select channels that may contain respiration and combine them. It can be repeated using a sliding window (e.g., a 64 second sliding window), for example, every few seconds (e.g., 5 seconds).

[0321] Active and passive envelope extraction can also be performed. Active signal envelope extraction can include the following steps: (i) invalid data insertion, (ii) outlier removal, (iii) IIR low-pass filter cutoff at 0.075 Hz (or some other low value), (iv) filter delay compensation, and (v) downsampling at 1 Hz.

[0322] Passive envelope extraction can involve Figure 8B The steps shown. This may involve any, some or all of the following steps: (i) channel combination, whereby the baseline of each band is removed by subtracting the minimum value over a sliding window, and the bands are combined by taking the average value over the bands. As shown, (ii) the process may, for example, extract the passive stream using (e.g., an anti-aliasing filter), after which (iv) downsampling may be performed. These steps, for example, reduce the original 48kHz signal by a factor (e.g., a factor of 3), thereby improving efficiency.

[0323] Optionally, to enhance detection of intensity SDB modulations, (iii) the spectrum may be spectrally denoised using a filter such as a Wiener filter (using a running SNR based on time variation), as well as respiratory channel selection prior to other processing.

[0324] The noise reduction process can be performed in the frequency domain. The purpose of this process is to finally obtain an estimate of the signal of interest only from the measured signal, that is, to remove the noise from the signal, because the measured signal also includes noise. This can be done by spectral multiplication (convolution in the time domain) of the noisy version of the signal (that is, the noisy version actually recorded by the system) and the spectral gain G (to be determined):

[0325]

[0326] in G(p, k) and X(p, k) are the kth components of a short audio frame p (1 / 16 second, 16 kHz, non-overlapping frames in the algorithm), corresponding to the estimated signal, gain, and measured signal plus noise, respectively. In the algorithm, the spectral signal X can be obtained using a 1024-point FFT applied to a Hanning windowed version of frame p. For example, frame p is obtained from the 16 kHz output of a decimated original 48 kHz signal from the microphone of a smart device.

[0327] The spectral gain G is a function of two components: the a posteriori signal-to-noise ratio and the a priori signal-to-noise ratio (SNR) of the audio frame. These terms refer to the calculation of the SNR with or without knowledge of the true noise for a particular audio frame p. They can be defined as:

[0328]

[0329] and

[0330]

[0331] Where S(p, k) and N(p, k) are the kth components of the short-term audio frame p of the signal and noise, and E is the expectation operator.

[0332] The gain G can then be described using an estimate of the above SNR value:

[0333]

[0334] Function g can take many different forms (see, for example: Ref. C. Plapous, C. Marro, P. Scalart, "Improved Signal-to-Noise Ratio Estimation for Speech Enhancement", IEEE Transactions on Audio, Speech and Language Processing, Vol. 14, No. 6, pp. 2098-2108, 2006). One example method that can be used is the Wiener filter.

[0335] (v) Once the noise reduction process is completed, the energy in the frequency band can be obtained from the noise-reduced spectrum signal and added to a first-in-first-out FIFO buffer (e.g., a 64-second circular buffer).

[0336] Signal quality and respiration rate

[0337] In some versions, the module 8812 may optionally implement signal analysis, for example, analysis of active and / or passive signals to estimate respiration rate and / or corresponding signal quality. In some versions, the respiration rate may be determined as described in International Patent Publication No. WO2015006364. Such information may be implemented as part of the absence / presence detection module to identify whether a person is near the sensing area, thereby improving robustness.

[0338] The system may calculate respiration rate separately by active and / or passive streams, and optionally combine these outputs based on the estimated signal quality of each stream.

[0339] Passive respiration rate

[0340] Optionally, respiration rate (active or passive) may be used, for example, when good signal quality is present, to improve the accuracy of the system. For situations where signal quality is poor, modulation may still be detected to identify clusters of apneas / hypopneas - even if real-time respiration rate is not available for parts of the night / sleep period.

[0341] In some versions, the submodule can be configured to process each of the above frequency bands (e.g., in a buffer) and estimate the respiration rate and / or signal quality (e.g., as described above) for each period (e.g., a 64 second buffer). This can be repeated every 5 seconds, consistent with the respiration rate determined from the active sensing signal (e.g., the respiration rate obtained by sonar).

[0342] Since the respiratory signal obtained by passive analysis is generally in the form of periodic pulses, some preprocessing is first applied to help improve the robustness of the algorithm. Exemplary steps may include any of the following:

[0343] 1. Apply a filter that matches the Gaussian impulse response

[0344] 2. Remove the baseline by subtracting the minimum value over the sliding window

[0345] 3. Limit to the 95th percentile to minimize artifacts caused by motion.

[0346] 4. Normalize on short sliding windows according to std.dev (standard deviation)

[0347] 5. Integrate the signal and high-pass filter

[0348] 6. Normalize again on the sliding window according to std.dev

[0349] This pre-processing is done to convert the pulse train into a signal more suitable for time domain analysis.

[0350] An autocorrelation can be performed on the signal and the first peak within the defined respiration range is selected as the respiration rate estimate. The peak convexity is calculated and used as our signal quality or confidence measure. This process can be repeated for each frequency band.

[0351] Finally, a respiration rate estimate may be calculated by taking the average of the four estimates with the highest signal quality. This signal quality output is simply the average of the highest four signal quality metrics. This output may optionally be output to the evaluation module 8820 or any of its submodules / methods.

[0352] 8.3.4.1.2 Evaluation Module 8820

[0353] like Figure 8 As shown, the evaluation module 8820 may include an SDB modulation evaluation submodule 8822 or method, a sleep staging submodule 8824 or method, and an SDB summary submodule 8826 or method. Each of these will be discussed in depth herein. In general, the SDB modulation evaluation process may perform SDB modulation evaluation (in frequency and / or intensity) for each signal stream from the extraction process (i.e., active signals, e.g., respiratory motion signals acquired by sonar or radar, and passive signals, e.g., audio acoustic signals) separately. The term "SDB modulation" may be understood as modulation of the amplitude modulation and long-term envelope changes of the respiratory / biomotion signal caused by SDB-related events.

[0354] In addition, the SDB modulation evaluation process can identify time periods (i.e., input signal portions) containing significant or high intensity SDB modulation (clusters) for each of the active signal stream and the passive audio signal stream respectively provided by the output of the extraction process. In addition, the SDB modulation evaluation submodule 8822 can generate a correction mask as previously described, which can provide sleep / wake characteristics for use as input to the sleep stage indication for improving the sleep stage process of submodule 8824. Figure 8 As shown and explained in more detail herein, evaluation of the SDB summary submodule produces an AHI and a binary marker, for example, which is output by optional output module 8830 (e.g., to a screen or other user interface).

[0355] 8.3.4.1.2.1 Evaluation of SDB Modulation Submodule 8822

[0356] The active (e.g., sonar) and passive audio streams (i.e., the signals produced by the extraction processing of the extraction module 8810) are processed separately to interfere with the spectral characteristics of the modulation in the quantized SDB frequency range (25 to 100 seconds). Many processing steps are similar for these two streams.

[0357] (A) SDB modulation in active (e.g., acquired from sonar) streams

[0358] For each I and Q channel representing respiratory effort, the following process determines the modulation intensity associated with a particular SDB modulation frequency. Fig.11 As shown, an envelope is extracted from each channel / signal. The envelope may be filtered and / or normalized. The envelope is processed by spectral analysis to determine the modulation frequency characteristics and / or its intensity characteristics within the SDB-related range (e.g., approximately 25 to 100 seconds). The modulation intensity at the SDB-related frequency may be output to the SDB summary submodule 8826, which will be discussed in more detail herein. Exemplary steps for generating such an output may include:

[0359] 1) Envelope extraction (FMCW extraction envelope):

[0360] a.NaN interpolation

[0361] b. Outlier removal

[0362] c. IIRLP filter with cutoff at 0.075Hz

[0363] d. Downsampling at 1 Hz

[0364] 2) Normalize the envelope by the mixer method (normalized envelope):

[0365] a. Initial scaling according to the overall median

[0366] b. Linearly combine the envelope of step a) above and the envelope scaled by the standard deviation on the sliding window

[0367] 3) Extract the spectrum characteristics within the SDB frequency range (obtain the modulation frequency and calculate the ibm (in-band metric value)):

[0368] a. After applying decomponentiation and spectral windowing, perform an FFT (e.g., 1024 points) every minute during a 6 minute window.

[0369] b. Extract the following spectral characteristics:

[0370] (i) fPeak: The frequency corresponding to the peak PSD within the SDB frequency range (ie, the SDB frequency band - e.g., 25 to 100 seconds).

[0371] (ii) ibm (in-band measure): the corresponding modulation strength or signal quality, written as:

[0372]

[0373] Where "P signal" is the power contained in a frequency band (e.g., narrow band) around the peak frequency, and "P noise" is the difference between the total power and "P signal"

[0374] Finally, an average of the two channels may be determined to produce a signal representing the modulation strength (average SDB in-band metric (ibm)). This average may then be evaluated in the SDB summary submodule 8826 to define SDB clusters (active clusters) for active streams (e.g., sonar or radar), which will be discussed in more detail herein.

[0375] (B) SDB modulation in passive audio stream

[0376] For the audio band channels (e.g., 9 channels in this example), the following process determines the modulation intensity associated with a particular SDB modulation frequency. Fig.13 As shown, an envelope is extracted from the combined channel / signal. The envelope may be filtered and / or normalized. The envelope is then processed by spectral analysis to determine the modulation frequency characteristics and / or its intensity characteristics in the SDB-related range (e.g., approximately 25 to 100 seconds). The modulation intensity at the SDB-related frequency may be output to the SDB summary submodule 8826, which will be discussed in more detail herein. Exemplary steps for generating such an output may include:

[0377] 1) Channel combination:

[0378] a. Remove the baseline of each frequency band by subtracting the minimum value over the sliding window.

[0379] b. Combine these frequency bands by taking the average value over these frequency bands.

[0380] 2) Envelope extraction:

[0381] a. Extract peak values ​​and apply a minimum peak separation of about 5 seconds

[0382] b. Interpolate at 1Hz and apply a mean filter to obtain the envelope.

[0383] 3) Envelope filtering, including integration and high-pass filter (passband frequency at 1 / 100 Hz):

[0384] This step is used to convert the pulse train (which is the signature of obstructive apnea in the passive audio signal envelope) into a signal more suitable for spectral analysis

[0385] 4) Envelope normalization - same as step 2) of sonar flow processing

[0386] 5) Extract spectral characteristics in the SDB frequency range - same as step 3) of active stream processing above.

[0387] This generated passive signal sensing based modulation strength signal at the SDB related frequency may also be output to the SDB summary submodule 8826 where it may then be evaluated to define SDB clusters (passive clusters) for the passive data streams.

[0388] (c) Calculate the sleep / wake correction mask for sleep staging in the active audio stream

[0389] Submodule 8822 may also generate a sleep / wake correction mask based on the generated cluster markers characterizing the presence of SDB clusters such as those detected in the active sensing stream. This approach may help address the following factors:

[0390] Ensure consistency between sleep staging (e.g. if performed based solely on sonar signatures) and SDB assessment

[0391] Active sensing signal (e.g., sonar) features for SDB screening outperform passive audio features alone

[0392] The mask is activated when the detected SDB clusters constitute more than a predetermined percentage (eg 20%) of the total recording period duration.

[0393] A mask is generated to mark the periods of high intensity SDB modulation to reflect that the periods should be marked as sleep periods. The output correction mask is input to the sleep staging module 8824, where the output correction mask is applied after the sleep / wake classification of the submodule, consistent with the SDB modulation determination in the mask, to correct the instances (periods) of the wake classification so that they are marked as sleep instances / periods.

[0394] 8.3.4.1.2.2 SDB Summary Submodule 8826

[0395] As previously described, the evaluation module can evaluate input signals (e.g., modulation strength, such as SDB-related frequencies from active and passive sensing signals) to determine an SDB classification flag, a probability of SDB risk, and / or a score for an SDB event. This determination can also be based on sleep stage information, such as from submodule 8824. This determination can initially involve the detection of SDB-related clusters that represent segments (time periods) that include SDB-related modulation. This determination is described in more detail herein.

[0396] (A) Identification of segments (clusters) with high intensity SDB modulation

[0397] Based on the modulation intensity metric of each active (e.g., sonar and / or radar) and passive audio stream, clusters of high intensity SDB modulation are identified. The method may include any of the following processes:

[0398] ··Construct a binary flag signal (time series) by evaluating a minimum threshold of metric values. For example, for values ​​of the strength signal at or above a threshold, the flag is set to true, otherwise it is set to false. The "strength" relates to the strength of the amplitude / power modulation and / or the strength of the frequency modulation of the active or passive sensing flow.

[0399] Optionally, the intensity signal may be filtered, for example by applying a 10-tap mean filter to the intensity signal.

[0400] The cluster flag signal may be obtained or modified by additionally evaluating an additional filtered intensity signal with a second threshold value. For example, for filtered intensity values ​​at or above the second threshold value, the flag is set to true, otherwise it is set to false.

[0401] This threshold can be optimized, for example, using empirical analysis, to maximize accuracy in detecting actual SDB clusters derived from PSG annotations. The cluster signal from this processing, Fig.11 (cluster from active sensing signals) and Fig.13 (cluster from passive sensing signals) is shown in the bottom graph.

[0402] (B) Classification

[0403] As previously described, the SDB summary submodule 8826 may include one or more classifiers to classify inputs from the above-mentioned processing for any of SDB risk determination, SDB identification and / or event scoring. For example, this module can calculate SDB-related features, such as one or more scalar values ​​that characterize the entire recording period, based on active sensing streams (e.g., any intensity signal, cluster signal, respiratory rate signal, signal quality signal, sleep stage, etc. from I channel and Q channel) and based on passive sensing streams (e.g., intensity signal, cluster signal, respiratory rate signal, signal quality signal, etc.). The module can, for example, apply a logistic model classifier to generate a binary SDB classification flag and an associated SDB risk probability, and / or apply a regression model classifier to generate (a) an estimated value of the AHI score or (b) a risk probability value that the number of apnea and / or hypopnea events included during sleep is greater than a threshold (e.g., a threshold indicating a sleep disorder). The module can also generate a sleep score adjustment (e.g., penalty) factor based on an evaluation of the SDB probability. This may be applied to modify the sleep-related score determined by the sleep staging module 8824 , which is input to the SDB summary module 8826 .

[0404] Exemplary SDB-related features for any of these classification processes (which may be determined by the processing device) may include any one, more, or all of the following features:

[0405] 1.cluster_ratio_active feature: This feature can be the proportion of the total sleep time composed of SDB clusters derived from active sensing streams (e.g., SONAR sensing). This feature can be derived by the processor as follows:

[0406] a. Evaluate the generated hypnogram data to obtain total sleep time and sleep wake mask. The hypnogram determined by the sleep staging process identifies different sleep stages (e.g., deep, light, REM) or wakes during the recording period. The sleep wake correction mask is an indicator time series for the recording period, which represents sleep or wakes in a specific time sub-interval of the time series. Therefore, the sleep indicator of the mask summarizes any of deep sleep, light sleep, or REM sleep, etc. as a sleep interval.

[0407] b. The sleep-wake mask is then evaluated based on its temporal relationship to the sonar cluster time series, enabling identification of specific SDB cluster-related times that occur with or coincide with sleep (ie, masked sleep indicative).

[0408] c. The proportion of features can be calculated by dividing the SDB cluster related time by the total sleep time (eg, SDB cluster time / total sleep time).

[0409] 2. cluster_ratio_passive feature: This feature is also the ratio of the total sleep time composed of SDB clusters as described above, but the processor that derives this feature has the SDB cluster time series determined from the passive audio stream instead of from the active sensory stream.

[0410] 3. fpeak_median feature: This feature is the peak intensity or peak median intensity that can be determined from the SDB modulation frequency time series from the active sensing stream (i.e., the intensity signal described previously). Optionally, this feature can alternatively, or as an additional feature, be additionally based on the passive sensing stream. This feature can be on the time series of modulation frequency from the sonar stream or the passive sound stream. This feature can be determined / calculated by the processor as follows:

[0411] From the active sensing stream (e.g., fpeak_median_active);

[0412] (a) Non-existent time periods can be filtered out;

[0413] (b) The I and Q channels, if present, may be combined for further processing, for example by taking the minimum value of the channel that represents the channel with the least noise; alternatively, the I and Q channels may be combined by applying a 3-point median filter to smooth the channels and extracting the value corresponding to the channel with the highest modulation intensity, and applying a 3-point median filter to smooth the ibm (in-band metric value).

[0414] (c) Then, for the period where the modulation intensity exceeds the threshold, the feature can be calculated by calculating the median of the signal or the minimum signal after combination. Optionally, the block of isolated one point can be excluded from the estimate. If the cumulative duration of the exceeded block is less than the threshold, the minimum frequency can be returned. If the active cluster ratio is large, the frequency feature can be calculated as the mean of the active cluster.

[0415] From passive sensing streams (e.g., fpeak_median_passive);

[0416] (a) A 3-point median filter can be applied to the modulation frequency time series for smoothing;

[0417] (b) The median of the passive cluster can be used as a feature.

[0418] A classifier may then evaluate any, multiple, or all of these features to generate a binary flag, a risk probability, and / or an estimated total number of SDB-related events (eg, an AHI estimate).

[0419] By applying a default logistic model to the classification features (e.g., cluster_ratio_active, cluster_ratio_passive and / or fpeak_median (e.g., fpeak_median_active and / or fpeak_median_passive)), the output risk probability and the output flag (e.g., positive SDB flag or SDB classification flag) output value are determined / calculated when the total number of apneas / hypopneas is greater than a threshold (e.g., AHI>=15).

[0420] An estimate of the AHI score (estimated sum) can also be calculated by applying a linear regression model to the same features. To avoid the risk of conflicting output values, the regression model can be "slaved" to the logistic model. Therefore, if the predictions of the regression model and the logistic model disagree, the AHI score can be modified based on the probability output.

[0421] The system can return an invalid output (NaN), e.g., confirming that an SDB risk determination cannot be made, under the following conditions: when the total presence time is less than a threshold (e.g., 4 hours); when the artifact detector rate exceeds a set threshold (e.g., indicating poor signal quality), and / or the default model returns a positive result (e.g., a positive SDB indication) and the passive feature-based model returns only negative results (e.g., there are conflicting results from different detection methods).

[0422] In some versions, the SDB threshold may be set to an AHI of 15. In this case, a positive result may indicate an AHI > 15. Following such a result, the user may be advised to see a doctor, for example, via an automated message on a display.

[0423] As previously described, the processor may also determine an adjustment factor or sleep score correction factor that may be used as a penalty to the sleep score, for example, when an SDB event interrupts sleep. Such an adjustment factor may be calculated by the processor according to the following function:

[0424] factor = min{1, slope×risk_prob+intercept}

[0425] The slope and intercept may be predetermined or set such that the factor satisfies the following two criteria:

[0426] (a) The factor will be equal to 1 (no penalty) for subjects classified as negative (e.g., SDB risk probability is 0 or below the threshold);

[0427] (b) When "risk_prob" or the SDB risk probability (eg, the probability that the user's AHI is greater than a total threshold (eg, 15)) is 1 (eg, 100% or greater than a percentage threshold), the factor will be equal to the minimum value (eg, 0.6).

[0428] Thus, in this example, the factor will vary between 1 (no penalty) and 0.6 (full penalty) relative to the determined value of the SDB risk probability. The factor can then be multiplied by any, multiple, or all of the following: (1) total sleep time, (2) deep sleep time, (3) rapid eye movement sleep time, and (4) light sleep time. As described in International Patent Publication No. WO2015006364, these sleep times can be determined, for example, based on motion signal analysis, whether the signal is from a sonar or radio frequency sensor. The factor can then be used to penalize the reported sleep score, physical relaxation score, and / or mental relaxation score, for example in the presence of a SDB probability risk. In some such versions, any sleep time adjusted by the factor can be output on a display, for example, relative to the unadjusted sleep time, to visually demonstrate the impact of SDB symptoms on the user's sleep quality or sleep time. For example, a bar graph of sleep time can be displayed on a display device with a specific color (or any other method), and, because of the penalty, a certain color or portion of the bar graph can be shaded or presented in a different color to show the penalty value relative to the whole. Alternatively, two adjacent bars (or other shapes) can be shown - one with a penalty factor and one without a penalty factor. Such a visual indication can motivate the user to seek treatment for SDB symptoms, or strengthen their willingness to follow advice, for example, to use a treatment device (e.g., an RPT device). Optionally, this penalized / deducted sleep time can be displayed only as an overall reduction, without drawing any boundaries to represent the penalty.

[0429] Corrected sleep-related scores (e.g., sleep score, mental relaxation score, physical relaxation score) based on SDB can be calculated as follows. The sleep score correction factor can be calculated as described above. The slope and intercept can be calculated such that: (i) for subjects classified as negative, the factor is equal to 1, (ii) for the case where the probability of AHI>=15 is 1, the factor is equal to a minimum value of 0.6, for example.

[0430] This factor can be used to reduce the components of the sleep-related score, including, for example, any or all of the following components:

[0431] Total sleep time

[0432] Deep sleep

[0433] Rapid eye movement sleep

[0434] Light sleep

[0435] The penalty therefore affects the sleep score, physical relaxation score, and mental relaxation score derived from these adjusted components. These scores can be determined according to the disclosure of patent application No. PCT / US2014 / 045814 filed on July 8, 2014, the entire disclosure of which is incorporated herein by reference.

[0436] 8.3.4.1.2.3 Cough assessment

[0437] In some versions of the present technology, cough assessment can be implemented, for example, using the extracted signals and / or other processing of active and / or passive sensing signals for the above sensors. Such a system can be implemented with a cough assessment module 8833, such as Fig. 8A For example, such an evaluation of generating an indication signal about a cough event can be implemented in combination with the above-mentioned evaluation module 8820, such as Fig. 8A As shown, or implemented as a separate module from the assessment module 8820. For a person who is sleeping (e.g., in bed, on a couch, in a chair), cough / wheeze / sneeze detection can be run in parallel with SDB detection. For when the person is awake, SDB detection does not need to be run because the input is always (or almost always) "awake". The sleep score can be reduced (penalized) as the cough worsens, for example, by manifesting as an increase in sleep interruptions or sleep fragmentation index. The overall quality of life score can be reduced by coughing or other respiratory events during the day and night.

[0438] In this regard, it is important to note that cough assessment can be used to evaluate a variety of cough types, some of which can be indicative of the severity and progression of a variety of serious respiratory conditions, as described below.

[0439] Asthma is a chronic inflammatory lung disease characterized by recurrent episodes of apnea, wheezing, and coughing (www.cdc.gov / asthma / interventions / background.htm).

[0440] Cough variant asthma is a specific type of asthma in which the main symptom is a dry, nonproductive cough, and there may be no other common asthma symptoms, such as shortness of breath or wheezing. This may be called a "chronic cough", can last for more than six to eight weeks, and can include sleep disruption.

[0441] A meta-analysis by Song et al. (European Respiratory Journal, 2015) estimated the epidemiological burden of chronic cough in the general adult population and found a high global prevalence (9.6%), comparable to that reported for asthma or chronic obstructive pulmonary disease. When the most stringent definition of “cough greater than 3 months” was used, chronic cough showed a prevalence of 7.9%.

[0442] Patients with cough can be divided into those with acute (usually viral) cough and those with chronic cough. Acute cough represents the largest single reason for consultation in primary care, while chronic cough is one of the most common presentations in respiratory medicine, a significant burden on the healthcare system.

[0443] It is of great significance for the attribution classification of cough, such as asthma, COPD, bronchitis, tuberculosis (TB), pneumonia, lung cancer, etc.

[0444] The social and economic burden of cough is substantial, especially when one considers the absence from work and school due to colds (estimated at 100 million people worldwide – Altman & Irwin, Cough: An Interdisciplinary Problem, 2010).

[0445] The worldwide cough treatment market is in the tens of billions of dollars; for example, recently, the global cold & flu oral syrup market was $11 billion, and the global OTC cough, cold and allergy medication market is estimated to increase at a CAGR of 4.9% from 2017 to 2022, reaching a value of $37 billion (Factmr, 2017).

[0446] The characteristics of a cough are related to its mechanism. The medullary "cough center" of the human brain prepares you to cough and activates the diaphragm downward to pull air into the chest cavity, the glottis closes, and then the diaphragm and intercostal muscles contract to build pressure in the chest cavity until the glottis opens and the pressurized air escapes (potentially with whatever stimulus caused the coughing action). Inhalation, forced exhalation, and glottis closure are sometimes referred to as the three stages of a cough, with the forced exhalation being the main source of sound that can be heard.

[0447] Thus, it can be seen that this inspiration and release is characterized by the detected chest wall movement (air is drawn into the chest cavity and then expelled under pressure) as well as the acoustic characteristics of inspiration, closing of the glottis, opening of the glottis, and release (indicating the "sounds" that one can associate with a cough). This has different characteristics than, for example, a wheeze, sneeze, gasp, crackle, or whistle.

[0448] Untreated flu or common cold (especially if accompanied by wheezing and coughing) can lead to life-threatening pneumonia. A person may also experience mucus of varying colors and volumes. An asthmatic cough may be productive (producing sputum) or nonproductive (dry cough) and may or may not be accompanied by wheezing (a high-pitched whistling sound caused by airway constriction). For example, a doctor may use auscultation to check for wheezing and crackles.

[0449] Most cough episodes are self-limited and disappear after the respiratory infection resolves with or without specific treatment. A cough that persists will require medical attention (for example, if it is associated with lung cancer). One of the most important things to do when managing a cough is to determine if an acute serious illness is causing the cough. The vast majority of acute coughs are caused by benign upper respiratory tract infections, such as the common cold (Chest Foundation, 2018).

[0450] Chan (European Respiratory Journal, 2010) noted that chronic cough (more common in women) may be the only presenting symptom in patients with obstructive sleep apnea. They also noted: "SDB cough is not well recognized by physicians, and some patients referred to specialist clinics for research cough may be misdiagnosed. Recognition of SDB cough is important because preliminary reports suggest that it responds well to SDB-specific treatments."

[0451] Therefore, it is suggested that an application-based cough detection system, such as the one described herein, may have various benefits to people, including in terms of increasing awareness of SDB in the population.

[0452] Other possible cough assessment solutions require the purchase and wearing of specific hardware / devices – such as wristbands, bracelets, necklaces, rings, patches, watches, or hardware / devices placed next to the bed / sleeping area or within the living space.

[0453] Many existing systems also require expensive external microphones, and these microphones may need to be placed on or very close to the body (e.g., as part of a lab-on-a-chip patch, etc., clipped to a lapel, placed on the neck), and used in very quiet environments. In addition, they may require large amounts of data storage. This may create privacy issues, as the session may also be recorded. Such systems also employ offline processing, including manual consideration of possible cough episodes. For body-worn devices, the battery capacity of the device ultimately limits the sampling rate / bit depth of the audio (i.e., reducing the potential quality of cough classification, and the discrimination between this and background / false sounds) and the time between recharges, and / or skin irritation due to adhesives when used for long periods of time; in contrast, non-contact systems can use a more powerful power supply or plug into an AC outlet / power supply via a transformer, etc. Previous technologies can also attempt to detect coughs or classify the cough type of a pre-detected cough, but require specialized hardware, such as by using a professional / specialized microphone placed near the neck with the above constraints. Such systems may wear and tear the microphone / transducer over time, be affected by handling, water / sweat, and come into contact with undesirable cosmetics. For example, they may be potentially undesirable in appearance and make the person appear uncomfortable to others. Therefore, they are not worn in public or replaced after washing.

[0454] In contrast, the cough assessment system described herein is based on non-contact (e.g., acoustic) cough sensing and is housed within an existing processing device as described herein (e.g., a smartphone or other processor-based device with access to any of the previously described active and / or passive sensing components). Thus, it can address the unmet need in understanding cough in a variety of settings, including nighttime settings.

[0455] As previously mentioned, this non-contact, "acoustic" type sensing can include sound, ultrasound, infrasound, and vibration related to mechanical waves in air, solids, etc. The "passive" sound processing described herein can be understood as digital signal processing and other techniques applied to electrical signals from one or more microphones (where the system "listens" to sound and determines whether coughing, wheezing, choking, snoring, etc. exist). This can include processing reflections and echoes from other sources. However, the previously described "active" processing involves sensing sounds that can be emitted by a system for motion detection, such as audible or inaudible sounds for sensing. For such sensing sounds in the range of about 18kHz to just below 24kHz, these can be implemented with ordinary speakers and microphones in smart devices (smart phone tablets, smart speakers, smart TVs, and radios, etc.) sampled at 48kHz. For higher sampling rates, higher frequencies can be used once the mechanical design of the speakers and microphones and the housing supports these configurations. The term sonar is used to describe such active systems (operating primarily in the air) that use / generate sensing frequencies above about 18kHz (which is the hearing threshold of most people). Ultrasound can be understood to start at about 20 kHz (above the threshold of hearing). Therefore, the system can actively operate by generating sensing sounds to include low frequency ultrasound.

[0456] Therefore, an example processing device or system of the present technology can provide easily available computers and audio processing hardware, such as devices that people carry and use daily, and can be located next to them at night, such as smart phones or smart speakers, smart TVs, smart watches, smart bracelets, smart necklaces or patches. The advantage of this system is that it can be operated by executing an application (e.g., a downloadable application (app) on any smart device (e.g., a smart phone or smart speaker)) in any operating system (e.g., Android or Apple). It can then supplement the above-mentioned non-contact sensor with a microphone (e.g., an integrated microphone, such as a microphone located at the bottom edge). The proposed system can operate by implementing a special mode that minimizes phone signal processing. For example, a multi-core CPU can be used to process audio samples obtained from a CODEC with a microphone in a frame. Typically, user-recognizable audio data is not stored in the phone's memory.

[0457] An example of such a system could track coughing episodes over time, categorize each coughing event or series of such events (e.g., a "dangerous" cough), and could recommend further testing or a medical appointment to the user to facilitate a timely and thorough medical evaluation. Such a device could help improve the user's quality of life.

[0458] The processing device can perform time-frequency audio signal processing on the real-time audio signal generated by the microphone and / or CODEC using an application program executed by the processor. Such a signal may contain information from the reflected audio signal (e.g., the sonar signal described previously), as well as the voice from the monitored person, and the background environment.

[0459] Optionally, cough detection (e.g., using a detection module or submodule) and cough classification (e.g., using a classification module or submodule) can be implemented as additional modules for use with, or receiving data from, active sensing related systems / modules (e.g., sonar or radar using a speaker and the same microphone to generate signals). Thus, cough detection and classification can be combined with or affected by, or provide an impact on: (a) data regarding whether a person is not present / is present in the sensing field (e.g., such information can be implemented to reject background noise, (b) data regarding sleep state (e.g., whether a cough event involves a recent arousal, and / or (c) isolation of the cough object (e.g., by considering or comparing the tracked / monitored breathing curve to determine whether a detected cough event is appropriately attributed to the monitored person and not to another person (e.g., a bed partner who is not being tracked / monitored).

[0460] Thus, the system can detect and / or classify specific features associated with breathing and related events (e.g., coughing features).

[0461] The analysis of one or more cough instances may involve (a) detecting one or both of (1) events (individual coughs) and (2) sequences of coughs (e.g., spasms of the same type), and (b) cough types (events or sequences) (e.g., productive or dry coughs). The system may also provide associated outputs to characterize causes. For example, the system may determine the probability that a cough is associated with any one or more specific diseases (e.g., COPD, asthma, gastroesophageal reflux disease (GERD), and upper airway cough syndrome, etc.). Such estimates may be generated, for example, in association with a computer model derived from empirical data including the analyzed scored cough types.

[0462] The system, for example using an application running on a processing device (e.g., a smart phone), can access different sensors to estimate physiological signals related to breathing conditions during sleep stages and wakefulness. In some cases, tracking of signals can be performed during detected sleep stages, including any of light sleep, deep sleep, and REM sleep. Therefore, the sleep stage can be evaluated in determining information about coughing (e.g., cough type and / or cause). For example, in REM sleep, sneezing would not be expected due to REM sleep disorders. Therefore, the event of coughing (rather than sneezing) may be a more likely sound event during REM sleep.

[0463] Additionally, the system can be configured to detect coughing during the day or at night.

[0464] For example, a cough can be characterized as occurring (or starting) in REM sleep (e.g., morning, evening) or in deep sleep. Therefore, the effect of a cough on sleep structure can also be detected. For example, a cough can be characterized by its effect on detected sleep (e.g., changes in sleep stages). In some cases, a specific wheezing pattern can also be detected, particularly during exhalation. During non-REM sleep, basal metabolic rate and ventilator drive decrease. For example, the system can be configured to determine a sleep score based on factors such as sleep quality and duration. Therefore, a cough assessment module can assess a user's coughing events based on a sleep score. For example, a sleep score may tend to decrease due to a cough (especially during frequent cough bursts).

[0465] Acoustic analysis can also be applied to estimate cough frequency, severity, and type of cough classification. In some cases, such analysis can be applied to detect specific wheezing patterns, especially when associated with detected exhalations.

[0466] Similar determinations can be made at night as well as during the day, during which coughing can lead to subjective and / or objective measures of fatigue and sleepiness being experienced. These can be associated with seasonal allergies to pollen (often in combination).

[0467] The system's microphone (e.g., associated with a smart phone or smart speaker, smart TV, or smart watch) can also be used to monitor and classify sound patterns consistent with chronic coughing or snoring, and separate these noises from other background noises such as cooling fans, road noise, etc. In some cases, low-frequency ultrasound sensing can optionally be applied to detect further information about the chest wall and body movements that may be attributable to coughing movements.

[0468] Several methods can be applied to detect and classify such cough-related features by the processing device.

[0469] In one example approach, manually processed features can be empirically determined using standard time, frequency, or time / frequency methods. A classification system can then be implemented using supervised learning. For example, a classifier system can be derived based on human marker events from, for example, PSG (polysomnography) sleep studies and acoustic sensing. Many PSGs already include audio recordings (sometimes combined with video recordings) that may be suitable for passive audio acoustic analysis. Ideally, PSGs will also include active analysis, where low-frequency ultrasound (e.g., above about 18kHz) is emitted and recorded, such as using sonar technology.

[0470] Such a classifier system can be based on linear or quadratic discriminant analysis, or a random forest classification model by generating multiple decision trees and using majority voting to improve the performance of the initial decision tree, or a simple neural network with one or more hidden layers and an output layer. Thus, the trained classifier can then be implemented for cough detection and type recognition (etc.) and the associated outputs as previously described. Shallow or deep neural networks can also be used for classification.

[0471] This approach has some advantages because the acoustic features have been explored in detail for speech recognition and the smart device processor can be optimized for such processing (e.g., computing the Fast Fourier Transform algorithm) (based on their telephone heritage, smart phones are particularly optimized for audio processing).

[0472] The system can be configured to detect characteristic patterns of snoring, sniffling, coughing or breathing difficulties, for example based on Figure 8CDetection may thus be achieved using techniques such as: (i) digitizing the microphone signal, (ii) a digital filter bank (an array of bandpass filters used to separate the signal into sub-bands), which are then processed in sequence using one or more (iii) time domain metrics such as energy levels and / or envelopes and / or zeros and / or kurtosis (particularly useful for identifying "bursts"), which may operate on each sub-band from the filter bank, as well as the full bandwidth signal content (iv) time frequency decomposition (e.g. using wavelets or short time Fourier transforms (STFTs), "bursty" noise searches (e.g. "cough signatures" using spectral analysis), and / or morphology (e.g. shape processing applied to the spectrogram)), which may operate on each sub-band or on the full bandwidth signal. Computing Mel Frequency Cepstral Coefficients (MFCCs), sub-bands may be selected that have narrower bands at lower frequencies and wider bands at higher frequencies. If sonar is running simultaneously, the sonar frequency band can be excluded from the cough processing, as well as any associated sonar harmonics (e.g., a FMCW 18-20kHz triangular scan at 16 scans per second can introduce a small artifact around 12kHz that should be considered during processing). These candidate cough events (and possible cough sequences) can optionally be cross-correlated with patterns identified in the sensed motion and / or respiratory data. By using these baseline aspects of the cough sound, and their optional relationship to the sonar-derived respiratory signal and motion detection, the system can achieve event, sequence, and spasm detection. A system with the above-described SDB processing can separate cough sequences from other modulated sound (audible) events such as SDB, PLM, RLS, and background noise with strong beats and / or repetitions, or indeed from other modulated events detected by the sonar-based movement and activity estimation subsystem.

[0473] Significant advantages of combining analysis of cough sounds with correlation to detected respiratory waveforms may include: (i) increasing confidence that a cough is actually present (e.g., by examining the cough in the audio band and making sonar estimates by assessing / determining the correlation between inspiration and forced expiration as seen in the waveform and the correlation between audible events), (ii) separating a partner's cough from the primary monitored subject, (iii) handling high background audio, such as when an elderly person turns on the TV loudly, (iv) rejecting speech and laughter (common confounding effects on cough detection), and (v) aiding in the separation of coughs (including types such as gurgling, sputum coughs, wet coughs (coughing up more than two tablespoons of sputum per day), and dry coughs, wheezes, whistles, sneezes, etc.).

[0474] The intensity of a cough can be estimated based on changes in the intensity sound level of the breathing waveform and / or the cough sound. When looking at the sound level, additional information about the intensity can be gathered by estimating the sensitivity of the speaker / microphone with a test playback sequence, and / or also by using an estimated range (distance) of the subject from the microphone based on sonar. In simpler terms, the system can estimate how loud / violent / intense the first and subsequent coughs are, even if the person moves relative to the sensing device (or vice versa).

[0475] In terms of cough detection parameters, examples of processing features of such manually processed signals may include calculation of the spectral content of the signal (e.g., by using a Fast Fourier Transform with peak determination (frequency domain), or time-frequency processing compared to processing using a discrete wavelet transform, appropriate basis selection, or peak determination). Remaining low frequency components identified in this processing may also be processed (e.g., separately) to determine longer time scale trends. Specific features for subsequent classification may be derived from such processing.

[0476] Respiration rate can also be monitored during coughing to understand correlations. An exacerbated cough can show an increased respiration rate during non-coughing periods relative to an individual's baseline, especially if the person is ill (whether it is a common cold, exacerbated symptoms, etc.). Frequent coughing at night may also affect sleep quality, and the presence of such coughing can be used to adjust the sleep score index.

[0477] Understanding cough frequency and severity / intensity as well as respiratory rate and body movements can also be used as an early predictor of asthma attacks (e.g., by detecting increased duration of cough trains, increased respiratory rate, alternating shallow and deep breaths).

[0478] For further background, a cough can be considered to include glottis opening (pop), followed by exhalation, and then glottis closing. A cough can be single (isolated cough), or a short sequence (a loud cough, followed quickly by a subsequent cough - usually lasting 5-10 seconds or longer). The envelope of a cough sequence can be calculated (or those very long sequences that are actually known to last for more than about 5 minutes). In other words, a cough can also be classified according to whether it occurs in a spasm (e.g., a spasm is generally considered to be a continuous cough longer than about 5 minutes). A cough can also be further classified according to its expectoration, or classified according to other terms, whether the cough is dry or wet (e.g., the presence of mucus indicates an expectorant cough). The exhalation time of an expectorant cough (e.g., in this system, detected by one or more measurement parameters based on an audio signal, measured based on a non-contact respiratory signal, measured based on chest movement) is usually longer than the exhalation time of a dry cough. Although it should be noted that there may be wide variation across the population (e.g., machine learning methods such as deep neural networks may be advantageously adapted to subject-specific snoring, or indeed when a subject's cough changes from wet to dry or vice versa over time).

[0479] Thus, any one or more of such cough-related classifications (e.g., (a) cough; (b) persistent cough / spasm; (c) expectoration rate (dry cough or wet cough, e.g., cough with mucus or with wheezing)) can be derived by classifying signals from sensors (active and / or passive) to classify.

[0480] A cough usually lasts at least 250 milliseconds, on the order of about 5-10 seconds. The inhalation and forced exhalation of a dry cough can produce a strong component up to 16kHz with most of the energy (near the 3dB point of the low-pass filter that can be used to suppress the sonar of 18kHz and above), with energy lasting from 0.1-4kHz, for example, 300 milliseconds, followed by a short quiet period, followed by other parts of the partial glottal closure / cough sequence, and so on, with a lower intensity component near 1.5kHz (e.g., "cough (loud), cough (weakened), cough (weakened), cough (loud)..." etc.). These are just examples, because age and gender, BMI, etc. can affect these ranges, where machine learning applied to large labeled (supervised) or unlabeled (unsupervised) cough-related signal data is important. Optionally, the classifier can also consider parameters such as age (e.g., children and adults cough can have different parameters) and gender.

[0481] For example, if a cough occurs during the night while the subject is trying to sleep, the cough may be detected by any one or combination of the following: (a) disturbances in the respiratory signal (e.g., in active sonar, gaps may be seen in the respiratory signal, and / or deep inspirations followed by rapid exhalations, or in passive respiratory analysis, significant disturbances in the spectrogram (particularly the initial "cough" sound is heard, (b) changes in heart rate (typically detected by the system as an increase from baseline heart rate within, e.g., the first 10 minutes, which persists for a period of several minutes following a cough primarily during exhalation), (c) an increase in blood pressure, and (d) gross (body) movement events (e.g., due to mechanical movement of the thorax). Overnight, a reduction in factors of the sleep score determined may be observed, particularly if a cough spasm occurs. Depending on the type of cough that occurs, a pattern of large inspiration followed by large exhalation followed by large inspiration may be seen (e.g., an increase of 50% relative to the most recent preceding breath). In contrast, other types of coughs may be accentuated by a large burst of exhalation followed by inspiration - a large expiratory rise time, a plateau, and then a recovery breath (or just a large exhalation).

[0482] The "spikes" induced in the respiratory waveform can result in an increased estimated respiratory rate, or in some cases a decreased respiratory rate over a short time scale. Therefore, in addition to estimating the respiratory rate, the waveform can be analyzed by processing local changes in respiratory amplitude changes via estimation of the envelope of the respiratory signal, and tracking various parts of the inspiration / expiration waveform morphology (specifically for identifying inspiration and forced exhalation followed by glottal closure).

[0483] Thus, audio processing may include applying a short-time Fourier transform (STFT) analysis to a sampled audio waveform obtained from an audio sensor, and estimating its normalized sub-band power levels. Mel-frequency cepstral coefficients (MFCCs) may be evaluated to distinguish snoring and / or coughing from speech. Spectral flux (analyzing the changes between power spectrum estimates) may be determined and evaluated to detect the onset of snoring. RMS (root mean square, or quadratic mean) may be determined and evaluated, for example in combination with STFT power levels and running a median filter, in order to distinguish cough-related sounds from background noise.

[0484] For deep learning models, the use of spectrograms is preferred, while for shallow models, MFCC spectrograms may be better.

[0485] Signal pre-processing (e.g., simple processing of the generated acoustic sound signal) may include applying a digital bandpass filter to keep the frequency content in the range of 100 Hz to 4000 Hz (or higher). This may be achieved by a direct form FIR filter using a Kaiser window or by other means. For example, if storage space limitations are a concern, the signal may be resampled to the desired frequency (e.g., 8000 Hz). Optionally, the signal may be companded, for example, by applying a μ-law compressor or similar method, or with a compandor. This may also be managed by other signal processing means, and companding is not necessary.

[0486] In the frequency domain (e.g., by using a transform of the signal output from the preprocessing loop), features for classification may include subband power (in Hz) in the following subbands: 0.01-500, 500-1000, 1000-1500, 1500-2000, 2000-2500, and above. For higher sampling rates, higher frequency bands may be considered. The bands may also be divided into smaller or larger segments. For example, a specific subband below 800 Hz may be considered, i.e., 0-800 Hz (de-meaned, removing the average value in the moving window). Other spectral measures that may be determined for deriving features may be the spectral centroid, as well as audio processing steps such as "pitch" - harmonic product spectrum, spectral flux, spectral spread, spectral autocorrelation coefficient, and / or spectral kurtosis.

[0487] In the time domain, features (derivable from the preprocessed output signal) may include zero crossing rate, autocorrelation coefficient and running amplitude measure. Other methods are to calculate short term energy and short time zero crossing rate.

[0488] In summary, exemplary features that may be derived by a processing device module for classification may include processing frames of audio samples (which may be non-overlapping) to determine:

[0489] ·frequency

[0490] Time including amplitude (can be used as a substitute for intensity)

[0491] Spectrogram

[0492] Wavelet

[0493] For an audio sampling rate of 48 kHz with 16 or 32 bit resolution, for example a frame size of 64 ms equals 750 samples.

[0494] Such frequency (FFT) and / or time / frequency (STFT, wavelet, spectrogram) features may include any one or more of the following:

[0495] Local peaks (measured peaks), and the ratio of the area of ​​the main peak (e.g., by integrating the area around the peak) to the surrounding peaks

[0496] Local and global maxima and any harmonics

[0497] The integral of the low frequency component (energy estimate), the integral of the high frequency component and / or the ratio of such low frequency and high frequency energy estimates, e.g. the ratio of low to high

[0498] Split into multiple frequency bands (e.g., using filter banks) and process the data within each sub-band to derive features as described elsewhere herein

[0499] Mel-Frequency Cepstral Coefficients (MFCC) - non-linearly spaced filter banks to approximate human hearing

[0500] Spectral flux

[0501] Spectral centroid

[0502] Harmonic product spectrum

[0503] Spectrum expansion

[0504] Spectral autocorrelation coefficient

[0505] Spectral kurtosis

[0506] Linear Predictive Coding (LPC)

[0507] Example temporal features may include any one or more of the following:

[0508] RMS (e.g., provides an estimate of the volume (loudness, e.g., a measure of dB) of each frame)

[0509] Zero crossing rate

[0510] Envelope (e.g., filtered absolute value of the Hibbert transform)

[0511] Pitch period based on autocorrelation function

[0512] Other techniques for the processing device may include applying a modified voice activity detection (VAD) process to reject speech, but retain respiratory events. In some versions, the processing device may employ a generic VAD to reject only low background noise, and retain sounds associated with speech and breathing. Speech recognition is a rapidly evolving field, and commercial cloud-connected services may be applied to mask speech / music to allow for improved SDB and cough detection.

[0513] Longer cough sequences can be processed to estimate changes in cough rate (e.g., coughs per time period (e.g., minutes or hours)) or to estimate modulation of cough rate during groups of coughs (e.g., coughs in each unified sequence or burst), for example, to estimate the shape of cough bursts (e.g., paroxysmal coughs that are frequent, and violent coughs that make a person breathless).

[0514] In the case where the breathing waveform has been estimated and is available to the processing device, it can determine therefrom any one or more other features for classification. Such waveforms and / or features can be derived as described in any of the previously mentioned references (e.g. PCT / EP2017 / 073613, PCT / US2014 / 045814, US-2018-0256069 and US2014-0163343). For example, any one or more of the following features can be determined and evaluated:

[0515] Inspiratory time and depth

[0516] Exhalation time and depth

[0517] Inspiratory-expiratory ratio

[0518] Notches / depressions in the respiratory signal waveform due to coughing (for example)

[0519] Long-term respiration rate (estimated from peak-to-peak, trough-to-trough, or zero crossings, or derived from spectral estimates)

[0520] The system can be configured to check (e.g., features for classification) whether there is a progression to full, sustained wakefulness associated with a cough spasm. The pattern can be evident from any one or any combination of non-contact sensing signals (e.g., RF, sonar, light signals), audio cough features, and cough repetition rate. Coughing during REM / deep sleep is rare (and coughing in REM is even less likely than in deep sleep), with most coughing occurring during wakefulness. When the system detects a possible cough feature on the audio processing side, if the monitored subject is determined to be in deep sleep or REM sleep, the probability that the cough feature can be determined to be actually a cough is reduced.

[0521] Of course, breathing-based REM detection can be affected by long sequences / cough spasms. As background, the breathing and motion-based sleep segmentation system used in this system is able to estimate the amplitude and duration of individual motion events as well as the amplitude and duration of each respiratory motion. The algorithm combines high-resolution estimates of these parameters into 30s epoch features, which are then evaluated to map the biomotion signal to different sleep stages. The presence of a bed is detected by increasing the received signal power. User wakefulness (W) is associated with higher levels and longer durations of detected motion and a higher degree of respiratory rate variability. Rapid eye movement sleep (REM) is associated with the highest variability in the detected respiratory parameters, while light sleep (intent corresponding to N1 and N2) and deep sleep (corresponding to N3) tend to increase the stability of the detected metrics.

[0522] Such sleep-related factors can also be implemented to enhance the detection of coughs from another object. For example, if the monitored object is determined to be in deep sleep or REM sleep based on a sleep stage determination such as a contactless sensor-based sleep staging module, it can be determined that the sound detected by the microphone or other acoustic sensor is from a bed partner. However, if the unmonitored bed partner has a cough spasm, this can result in the awakening of the monitored object. In this case, the detected respiratory waveform can be evaluated via contactless sensing to verify the source of the cough (e.g., monitored object, bed partner) using processing logic, such as by detecting that coughing followed by awakening is significantly more common than coughing during sleep and then awakening. In the detection of the first case, the system can determine that the cough is attributable to the monitored object, while in the second case, the system can determine that the cough is attributable to other situations.

[0523] When the incidence of coughing increases at night, the sleep score of the monitored subject may decrease. During the day, the recent activity level and the cough scores of the previous day and the last night can be used as inputs for risk classification of exercise-induced events (e.g., asthma attacks or COPD exacerbations), which can be improved by such inputs. Analysis of the decrease in sleep score caused by coughing (with arousal) or snoring, as well as the increase in resting breathing rate and / or changes in the inhalation / expiration ratio can indicate a worsening respiratory infection, and in turn indicate an increased risk of worsening of the subject's condition, such as due to exercise-induced events. In other words, when the user becomes more breathless at night and during the day, this is seen in changes in breathing (including coughing) parameters and in the decrease in activity / mobility when the person becomes more uncomfortable.

[0524] The duration and severity of coughing are monitored at night, providing feedback that allows quantification of the beneficial effects (or other) of any proposed treatment. Various exercise regimens, treatments or recommended changes (medication (adjusted amount of a given medication), behavior (deep breathing, avoidance of cold air, etc.), or other) may be offered to the patient. A worsening of the patient's cough at night and / or during the day may indicate a need to adjust or cancel the proposed treatment. Various parameters may be used to estimate the patient's worsening cough, such as the total duration of the night cough, the number of cough spasms, the severity of each spasm, etc. The quantitative criteria for "worsening" may be stricter than the quantitative criteria for "improvement". In addition, the quantitative criteria for "worsening" may be adjustable to become more stringent as the patient's health and / or cough worsens.

[0525] In a bedroom environment, the processing device can generate recommendations about suitable sleeping conditions - such as in terms of optimizing sleep quality, and in terms of reducing the risk of coughing and ultimately deterioration. This can be achieved by monitoring and / or controlling any one or combination of air temperature, humidity and air quality. These available parameters can be automatically controlled via a building, air handling or local humidifier and / or air conditioning / heating unit, which has a suitable controller or control device such as combined with the processing device. During the day when the bedroom is empty, these systems can be provided with a controller to increase airflow and automatically open vents / blinds to reduce humidity and freshen the room. Such settings and / or controls can be based on a processing device that detects cough events and / or types as described herein.

[0526] The system can also manage (e.g., using a reminder process or module) how long bedding (e.g., sheets, blankets, etc.) has been on the bed (e.g., reminding the user to wash the bedding weekly, or automatically sending a warning to a caregiver, etc.), as well as provide recommendations about the temperature at which to wash them (e.g., to reduce dust mites, germs, etc.) or other recommendations (e.g., adding additional laundry detergent at a lower temperature, or washing at 60°C supported by the bedding). The room can also be automatically heated to a suitable sleeping temperature when bedtime is approaching. The system can also provide specific feedback suggestions to the user, such as adjusting the humidity in the room based on measurements taken and / or analysis of local weather patterns. Control of such adjustments can also be based on a processing device that detects cough events and / or types as described herein.

[0527] As an alternative to descriptive and predictive supervised machine learning with handcrafted features, another form of machine learning called deep learning can be implemented. This typically relies on larger amounts of scored (labeled) data (such as hundreds of nights of scored sleep data and / or daytime cough data). This approach can implement many interconnected layers of neurons to form a neural network (deeper than a simple neural network), so that each layer "learns" increasingly complex features. Machine learning can use many more variables than human handcrafted features.

[0528] Convolutional Neural Networks (CNNs) are widely used in image processing to infer information - such as for facial recognition, and can perform well in this context (all inputs can be assumed to be independent of each other - such as pixels from one object are independent of each other). These can also be applied to audio spectrograms, but do suffer from the fact that the frequencies observed in the spectrogram may belong to many cumulative sounds - or indeed complex constructive and destructive interference (e.g., phase cancellation). Where possible, the system can also assess the periodicity of the audio. In effect, the system cognitively "learns" the temporal characteristics from the intensity, spectrum, and statistical estimates of the digitized audio from the recording device. This representation is not speech - but a different breathing problem (but can benefit from the speech and sound processing found in modern smart devices).

[0529] In contrast to CNNs, not all problems can be represented as problems with fixed-length inputs and outputs. For example, processing breathing sounds has similarities to speech recognition and time series prediction, and can therefore benefit from a system for storing and using contextual information such as a recurrent neural network (RNN), which can take previous outputs or hidden states as input. In other words, they can be multi-layer neural networks capable of storing information in context nodes. RNNs allow for processing variable-length inputs and outputs by maintaining state information across time steps, and can include LSTMs (Long Short-Term Memory type "neurons" to enable RNNs to increase control over them) to manage the vanishing gradient problem and / or by using gradient clipping.

[0530] In summary, core processing, such as in one or more modules of a processing device (or having one or more processors in communication with a processing device), may include any one or more of the following:

[0531] Ultrasonic detection of breathing and movement. This can use frequencies of around 15-20kHz or higher

[0532] Audio detection and classification of respiratory events, using microphones to detect at standard audible frequencies from 100Hz to 5kHz or higher. This audio processing can be fused with breathing patterns and movements derived from ultrasound detection

[0533] Optionally track activity and steps during the day using a smartphone accelerometer and / or gyroscope or other motion or other sensor (PPG) or service, and optionally correlate activity or step counts or changes in activity or step counts with cough events / types

[0534] Personalized AI insights using machine learning features and deep learning applied to large field datasets

[0535] Thus, based on acoustic / sound processing and / or motion processing, such as from sonar and / or radar sensing, processing may be implemented for detecting different types of coughs, wheezes and gasps in bed while awake (particularly after getting into bed, and after waking up in the morning, as coughs may be exacerbated by changes in body position). Thus, coughs may be classified relative to detected body position, such as changes in body position, for example to check whether a cough subsides when in a prone versus a supine position (or vice versa).

[0536] This detection can also be implemented in association with a sleep staging system that determines / indicates that a person is asleep. Thus, the system can evaluate sleep-related information such as previously discussed for detecting coughing.

[0537] Additionally, such a system may also enable cough detection in a more mobile or passive arrangement without the use of active sonar and its use of speakers, such as using only microphones to detect coughs when a person is moving (e.g., if a processing device such as controlled by a user activates microphone-related sound sensing when the mobile processing device is moving (i.e., not in a fixed location such as on a nightstand).

[0538] In situations where both active (e.g., ultrasound) and passive detection are available (such as in a bedroom or vehicle), the processing device can measure / detect biometrics of a specific user (the person being monitored). For example, such as International Patent Application No. PCT / EP2016 / 058789, the entire disclosure of which is incorporated herein by reference. Using the biometrics so determined, a detected cough can be associated with the person being monitored. For example, if a bed partner coughs, such acoustic events can be excluded from affecting the cough measure of the monitored person determined by the processing device.

[0539] In cases where both accelerometers (and / or gyroscopes) and passive detection can be implemented and made available (such as during daytime activities when the phone is in a pocket), the processing device can then associate a cough (detected acoustic cough event / type) with that person (rather than another person or background noise) by evaluating the motion signal from the accelerometer of the phone in the pocket that detects chest movement associated with a cough, such as by evaluating the temporal correlation of the sound (cough sound) with the necessarily associated motion (chest / body movement associated with the cough).

[0540] A useful feature is to correlate cough intensity (which can be a parameter based on loudness and duration, which can be corrected for range (e.g., adjusted or normalized) if distance metrics are available from sonar) and frequency (and its order, ranks, spasms) with respiratory rate variations over longer time scales, and thereby estimate the risk of asthma or COPD exacerbation based on a personalized estimated snoring metric.

[0541] Other aspects are detection of wheezes, dry "hacker" cough signatures, etc. Chronic coughs with and without wheezing episodes (e.g. extra "tones" in the spectrogram in the range of 0.1-15) can also be detected.

[0542] It can be seen that by detecting sleep stages, snoring and coughing, the processing device can associate these events with sleeping difficulties (ie, coughing or other restlessness or snoring).

[0543] Analyzing cough parameters over multiple days and nights can also provide insight into underlying asthma, as coughs caused by asthma may be worse at night. For example, the presence of a productive cough may increase the risk of COPD exacerbations.

[0544] In order to manage and / or assess the COPD condition, the processing device may detect dyspnea (shallower breathing), such as by assessing changes in the inspiratory / expiratory ratio (airflow limitation during the expiratory phase, which causes prolonged exhalation in airway obstructive diseases-one of the indications of COPD) and increases in respiratory rate, such as changes in long-term respiratory rate variability assessed via adjustments on a longer time scale (e.g., intra-night or night-time changes). This may be considered in relation to (such as by comparison to) group normative values, as well as compared to individual baseline values. For such COPD assessments, the processing device may also detect and assess increases in cough severity (repetition rate, intensity, duration, etc.) to indicate changes in the COPD condition.

[0545] A system of two processing devices (e.g., two smartphones in a room or a smartphone and a smart intercom) can be implemented to separate coughs from near and far users. These devices can transmit parameters (e.g., cough-related parameters) over a network (e.g., the Internet or short-range wireless (such as Bluetooth, Wi-Fi, etc.)), such as to help attribute cough events to specific people being monitored.

[0546] As previously described, the processing device may track the inspiratory and expiratory waveforms using active and / or passive acoustic processing.

[0547] Acoustic processing is designed to be able to handle noisy environments, such as when using a television, tablet or other smartphone audio (phone / video calls), speech, road noise, etc. For example, the processing device may filter its acoustic signal to remove frequencies associated with such sounds.

[0548] Depending on the type of neural network used, local phone artificial intelligence (AI) acceleration can be used to reduce power consumption (and can benefit, for example, CNN-type approaches commonly used for image processing, or RNNs more commonly used for audio processing).

[0549] In some cases, the processing device can control or provide output related to drug use. For example, such as by comparing the cough-related data before using this drug with similar data after / during using this drug, it can provide instructions about whether the user is helped by cough suppressants, antibiotics. Such drug use information / data can be input into the processing device to provide correlations such as use time, drug amount, drug brand, etc. The processing device can then further characterize or provide any changes in cough properties and whether it is relieved by drugs. In addition, such cough events can be used to identify whether they respond (change) relative to the use of other non-cough drugs (such as asthma drugs). If so, such coughs can be identified as coughs related to or not related to asthma. Similarly, if the cough events detected in this way still exist after the user stops taking the drug (such as after the asthma drug and / or inhibitor or antibiotic is discontinued). The system may identify such persistent coughs by warnings to further consider other medical problems, such as lung cancer-related coughs.

[0550] The technology enables rich and detailed chronic disease screening, analysis, and management, such as on a consumer / patient's smartphone or smart speaker, without the need to inquire about the purchase of additional hardware. Patients can be empowered, such as by simply downloading an app, to understand and, to some extent, manage their health, using smart apps that enable new sensing capabilities to gain insights into health data on their smart devices, and optionally through connectivity (processing assistance) through cloud software services.

[0551] about Fig. 8A Examples of such a system of cough assessment module 8833 may be further described with respect to Figure 8B Such a module may be included in Fig. 8Aor 8B system or implemented separately. The system may include an active sensing module 8902 and a passive sensing module 8904. The active sensing module may generate a sonar-based sensing signal via a speaker and receive a reflected version via a microphone and perform demodulation (e.g., mixing) to generate a raw motion signal. The passive analysis may perform filtering and / or other and signal conditioning (e.g., preprocessing). The system may include a sleep segmentation processing module 8906, such as for signal extraction and feature extraction. In the evaluation module 8908, the respiratory rate, signal quality, and motion / activity counts from the active stream may be determined based on the epoch and the raw motion signal. Similar generation of such signals / parameters may be derived from the passive stream in the process / module 8912. Since one or the other processing stream may be affected differently by the nose / interference / external movement, etc., the signal quality of both may be achieved to weight the features. Features such as for snoring classification and / or fingerprint recognition (e.g., cough features) may be derived from the passive signal stream in the feature extraction process / module 8914. Machine learning features may also be extracted in module 8916 for such classification. Thus, using such features, the snoring classification process / module 8920 and the cough-related fingerprint process / module 8918 may classify the passive stream to produce output 8928, such as cough events, snoring, wheezing, panting, etc., respectively. Module 8910 may process parameters from module 8908 and raw motion signals from active stream processing at 8902 to determine respiratory effort, such as a respiratory effort signal. Using such signals and sleep segmentation information from the sleep segmentation process / module 8906, the SDB assessment process / module 8922 may determine SDB-related outputs, for example, by classification, such as in the manner previously described. Using such assessment, including sleep segmentation information, sleep stages may be output in process / module 8924, and sleep scores may be output in process / module 8926. Additional output features in 8928, such as AHI, SDB events, signal quality, etc., may be output based on the SDB assessment.

[0552] Snoring Detection Example

[0553] One approach to implementing a snore detector is to consider real-time or near real-time processing (e.g., updating features twice per second) and / or "offline" operations at the end of the night / period, such as post-processing, which may include snore classification processing. In this case, "offline" may be understood to mean processing the data at some delayed point in time (e.g., after a sleep period). Whereas "online" processes may be understood to involve real-time or near real-time processing. The fact that the data does not have to be processed in real-time or near real-time allows the data to be transmitted to a remote server (or other cloud computing device) to enable the use of more or greater processing power. This facilitates the processing of more data over a longer period of time, which may also allow long-term trends in the data to be identified. Examples of the present technology may use local and / or remote processing to implement one or both of such online and offline processing methods.

[0554] For example, near real-time snoring processing may include filtering (such as noise reduction and / or removal of active sonar waveforms - e.g., low pass filtering to remove components above 17kHz - such as FIR anti-aliasing filters, and / or notch filtering to remove power supply electrical sound at 50Hz or 60Hz [depending on the local area power supply frequency]) and down sampling (to reduce data rate and processing requirements - e.g., resampling 48kHz audio to 16kHz audio). Notch (such as IIR) or other adaptive filtering may be used to suppress artifacts that may be introduced when a smart device (such as a cell phone) is charging while recording audio, as the charging cycle may introduce audio artifacts.

[0555] For example, an audio classification module for snoring may compute any one or more of the following three outputs (e.g., metrics):

[0556] The probability of snoring at a given second [p snore]. Such a probability can be calculated as described in more detail herein. This can be calculated by a neural network (e.g., a deep neural network DNN, a convolutional neural network (CNN), a recurrent neural network (RNN), etc.). This value can be used for processing and can also be displayed as a snoring indicator.

[0557] The root mean square (RMS) of the audio signal for a given second [Audio Rms] (eg, the input audio may be normalized by the RMS of the current second). This may represent an estimate of the signal strength.

[0558] Summary value for a given second of the representation of the output audio [AudioRepRaw]. This can be a relative strength measure that can be used to plot the relative amplitude of the signal. This summary measure is Fig.36 These metric outputs can be evaluated by an offline classification module to form additional outputs of the snoring detector (see, e.g. Figures 34 to 36For example, the total snoring time may be determined based on the classification and may be displayed in Fig.36 The user interface example in FIG. 1 shows a "snoring clip". As shown in the figure, the play button (see Fig.36 A play arrow symbol ) is drawn near the user interface to allow each of multiple audio signal segments or sub-segments of a larger audio period that can be selected for display based on the classification to allow the associated audio signal to be played back using a speaker of a device presenting the user interface.

[0559] For example, the processing may include, for example, calculating Mel-frequency cepstral coefficients MFCCs, and then using a CNN or RNN model (such as implemented in a machine learning platform (e.g., Tensorflow)) to estimate the probability of snoring at a certain frequency (e.g., 2 Hz), and to estimate the relative intensity of snoring at, for example, another frequency (e.g., 1 Hz). The average RMS may also be calculated for display purposes, for example, as an approximate representation of the intensity of snoring in a graph. Such a graph is shown in FIG. Fig.36 The system can also mark or select periods of high probability and / or high intensity snoring for user playback (see, e.g. Fig.36 ).

[0560] In one example, MFCCs can be calculated from filtered audio (e.g., 16kHz); for example, 15 MFCCs are calculated for every 25ms of audio, with a step size of 10ms, leaving 98 parts to calculate MFCCs. To remove the DC component from the MFCCs, the first MFCC can be deleted. This makes the size of the final MFCC matrix 14×98 for a given second.

[0561] In one implementation, the Tensorflow module (or some other implementation of an AI model that has been trained on a large labeled dataset of snoring and non-snoring audio) is called twice per second to provide a 2Hz output. This is done to reduce errors when snoring starts near the end of the current window. In order to call the model twice, a portion of each previous call to the process (e.g., the last half-second value of the MFCC) is retained in the memory of the process representing its state. The model is then called twice: (i) half a second of the MFCC in the state and the first half second of the current MFCC, and (ii) a full second of the current MFCC. The last half second of the current MFCC is then stored in the state for the next function call to the module. Before each call to the Tensor flow module, the MFCC is expanded into a vector of length 1372. The vector is then scaled before the call. Each call to the Tensor flow module produces a single output p snore value, which is stored for use and / or output. Thus, for a period of audio, processing (which may be local and / or remote, online or offline) may be implemented to produce a time series of pSnore values ​​associated with the time of the audio signal.

[0562] An exemplary offline version of the snore classifier may be used to post-process data collected from the snore audio classification model (e.g., using psnoring values) and compute snore outputs that may be displayed in a user interface. During a period or group of periods (if there is more than one period during a night), the offline snore classifier may be implemented to generate additional outputs (e.g., four) for an end user, e.g., in an application programming interface (API) format, including any one, more, or all of the following: (i) Total snoring duration [snoringDuration]—a count of the total amount of time the user spent snoring, such as during a particular sleep session or group of periods (e.g., a sleep night with several interruptions during a night) (ii) Snoring flag [snoringflag]—a binary flag indicating when the user snored, (iii) Snoring event times [snoringsnap]—a list of the number of times the user snored, which may be used for audio playback, (iv) Audio display signal [snoringDispSig]—a simple representation of the amplitude of the audio signal recorded per second (e.g., an amplitude metric based on the signal standard deviation). (e.g., for user interface (UI) purposes).

[0563] The snore classifier algorithm may perform any of the following processing operations: (i) post-processing psnoring values, (ii) thresholding psnoring to create snore markers, (iii) snore event time detector, (iv) blocking snore markers into chunks (e.g., 5-minute chunks), and / or (v) post-processing the audio display signal. Each of these operations is described in more detail below.

[0564] (i) Post-processing pSnoring values: As mentioned above, pSnoring looks at the short term of the audio signal (e.g., only a given second) and does not represent knowledge of long term trends in the data. Therefore, it is useful to post-process pSnoring into a single feature that is more robust to interference. A logistic regression model can be implemented with, for example, multiple input features (e.g., 3). The model can have two features derived from pSnoring and another feature derived from the audio RMS. The first of the two pSnoring based features can be a percentile filtered version of pSnoring with different window lengths. This is performed to assess snoring levels over larger time windows, as pSnoring typically assesses short term data (e.g., one second of data). Another feature is a signal quality based feature that is derived from an estimate of the breathing rate, e.g., between 5 and 35 BPM, which in turn is derived from the audio RMS. On a common time scale as a sleep structure map, pSnoring is a function of the breathing rate of the sleep signal. Fig.38 A display of the time-varying respiration rate in the signal is shown in . This feature increases robustness to non-periodic noise sources that are not snoring, but that may otherwise confuse the neural network. The output of the model is a measure of the likelihood of snoring (hereafter referred to as pSnoringFilt) and can be used internally by the following operations of the module.

[0565] (ii) Thresholding to Create Snoring Labels: For this analysis, snoring labels are created by thresholding pSnoringFilt from a previous process (e.g., repeatedly generated at a 2 Hz frequency). The process can also use the hypnomorphogram to eliminate detected snoring portions where the sleep staging algorithm has provided an absence, wakefulness, or interruption state for more than a certain amount (e.g., three minutes). In other words, the system can recognize that the detected snoring is not from the monitored user because the sleep system has recognized that the monitored person is awake, absent, or otherwise interrupted from sleep. When the system is designed to detect only the sleep and snoring of the nearest person, this assessment can help reject the sound of a bed partner's snoring.

[0566] (iii) Detector of snoring event times / segments: In order to provide the user with multiple snoring audio segments to listen to, the algorithm detects up to a certain number (e.g., 5) of the most likely snoring events and provides timestamps for the start and end of each audio segment. Each segment has a maximum length (e.g., one minute). This example uses Fig.36The algorithm can be explained by referring to "snore segments" of the audio. When pSnoreFilt is equal to or greater than a threshold, a snoring segment or segment is extracted from the segment. This threshold can be significantly higher than the threshold used for more general calculations of nighttime snoring percentages. If it is less than the maximum number of available segments, the algorithm can report as many available segments as possible. For example, given a recording with very little snoring, the user may only get three snoring segments from the algorithm. In the case where the snoring flag is not set, pSnoreFilt is set to 0. Before identifying the highest peak in pSnoreFilt at a minimum interval of 5 minutes, the average of the pSnoreFilt values ​​can be determined over a sliding window (e.g., 60 seconds). The algorithm then selects the highest valid peaks (e.g., 5) and returns a window + / - 30 seconds for each audio segment.

[0567] (iv) Blocking snoring markers into chunks: In one implementation, the snoring status may be reported in a more granular form, such as providing output in chunks on the order of minutes (e.g., 5 minutes). As a result, the higher frequency (e.g., 2 Hz) snoring markers are blocked into chunks (e.g., each 5 minute long) before the final snoring marker is output to the user. If more than 50% of the 5 minute snoring marker chunk has been marked as snoring (taking into account the higher frequency (2 Hz) snoring markers), the entire chunk is marked as snoring. The output snoring markers from this processing are then upsampled to be displayed in a thirty second (30s) period. The final snoring marker may be provided as an API output. A total snoring duration output is calculated from the long term snoring markers.

[0568] (v) Post-processing the audio display signal: The last part of the offline snore detector is a brief post-processing of the audio display signal. A standard deviation filter with a 5 minute window is applied to the entire display signal before normalizing it between 0 and 1. The output of this part is the final audio display signal. An example of this audio signal is shown in Fig.36 The middle panel of the user interface display is shown as FIG.

[0569] Estimated flow limitation / breathing-effort:

[0570] RF, sonar, and / or passive acoustic analysis can be used to estimate flow limitation based on an estimate of respiratory effort. Specifically, this involves the process of using such sensing technology to estimate the presence or degree of flow limitation (or some other respiratory effort metric). An example can output a sign or indicator (absolute or relative) indicating flow limitation / breathing-effort.

[0571] Flow Limit (FL):

[0572] May occur when esophageal pressure increases are not accompanied by increases in flow and is a noninvasive assessment of residual upper airway flow limitation

[0573] Associated with negative health outcomes and can disrupt sleep

[0574] Can occur chronically, i.e. may not be associated with apnea / hypopnea / desaturation events

[0575] Can be used to monitor COPD (expiratory FL)

[0576] Usually manifests as a flattening of the flow signal morphology

[0577] An exemplary process for estimating flow restriction / effort using passive acoustic processing may include:

[0578] Time / frequency methods, such as CNN / MFCC on spectrograms, cepstrum, short-time Fourier transform (STFT);

[0579] Pathways using active sonar, RF, light detection and ranging (lidar) to estimate flow restrictions / efforts, other optical pathways may include:

[0580] Inference from biomotion signal morphology (reference point analysis, skewness, kurtosis, respiratory partial area, etc.)

[0581] This will typically be looking for a metric that describes some kind of waveform flattening (taking into account CW / FMCW distortion where appropriate)

[0582] Doppler radar can be used to infer the inspiratory / expiratory portion of the breathing cycle (when available)

[0583] Pathways using passive acoustic and active sonar, RF, LiDAR, other optical methods to estimate flow restrictions / efforts can include:

[0584] Using passive and active features together can provide synergistic improvements to the contactless sensing modality, such as by increasing confidence, and allowing different types of flow restrictions to be captured. For example, a flow restriction may correspond to a less severe detection threshold than a full occlusion event, so different sensing techniques may allow different detection thresholds to be used.

[0585] These estimates can be combined with arousal / activity measurements to estimate respiratory effort related arousal (RERA). One potential use case is to recommend different therapy (e.g., MRD) for a group with low / mild AHI but flow limitation. It can also be implemented as part of sleep score deterioration (e.g., reduce sleep score based on the number and / or time period of flow limitation events).

[0586] The processing can also estimate the location / type of obstacles by analyzing the characteristics of the audio signal.

[0587] Separation of obstructive, central, and hypopneic events

[0588] The following event context can serve as a guideline for event classification:

[0589] (a) Apnea may be considered: a ≥10 second (optionally plus or minus 2 seconds) decrease in pre-event baseline with a peak signal shift of ≥90% (optionally plus or minus 5% or 10%) using oronasal thermal signals, PAP device flow, or the optional apnea sensor. No desaturation or arousal is required.

[0590] (b) Obstructive apnea can be considered as: the apnea described above in (a) above, in which respiratory effort continues when respiratory arrest begins (and, possibly, paradoxical breathing throughout the apnea);

[0591] (c) Central apnea can be considered: there is no respiratory effort in the above apnea (a).

[0592] (d) Hypopnea may be considered: a peak signal excursion ≥ 30% of the pre-event baseline for ≥ 10 seconds using nasal pressure, PAP device flow, or the optional hypopnea sensor, and there is a ≥ 3% oxygen desaturation from the pre-event baseline or the event is associated with arousal. The system may assess: (i) snoring during the event, (ii) increased inspiratory applanation of the airflow signal compared to baseline, and (iii) chest-abdominal discrepancies occurring during the event.

[0593] (e) Obstructive hypopnea may be considered: hypopnea (d) where - if any of (i), (ii) or (iii) is assessed as present

[0594] (f) Central hypopnea may be considered: Hypopnea (d) - if none of (i), (ii), or (iii) is assessed as present (i.e., (i), (ii), and (iii) are absent;

[0595] (g) Central apnea can be considered: ≥3 consecutive central apneas and / or hypopneas separated by crescent / decrescendo changes in respiratory amplitude with a cycle length of ≥40 seconds, and ≥5 central apneas and / or central hypopneas per hour of sleep associated with a crescent / decrescendo breathing pattern recorded during a monitoring period of ≥2 hours.

[0596] Obstructive sleep apnea in active passive analysis:

[0597] Obstructive apneas can be classified using only passive acoustic streams, as such events can be represented as periods of silence that last for periods close to the time of intense snoring. The periodicity of the events can be used for detection. This signature is very clear for severe cases, and has a different signature for moderate events where snoring, wheezing sounds may be present in the actual event.

[0598] The most notable feature of an activity signal (eg, sonar, RF, optical, etc.) relative to an obstructive apnea is a decrease in amplitude. For example, many events demonstrate a sustained respiratory effort (eg, respiratory motion) through an apnea event.

[0599] Active signals can be extracted in the process of combining distance bins using principal component analysis techniques. This can be supplemented by detecting characteristic snoring and inhalation sounds that can be extracted from passive signals. The classification system can output a sleep apnea classification, for example, by evaluating the count of events to grade moderate or severe obstructive sleep apnea (OSA). For example, for subjects with an AHI range of about 15 to 30 events (optionally plus or minus 2 or 3 events at one or two range limits), moderate OSA can be identified. Similarly, severe OSA of subjects with AHI above the moderate range (e.g., above 30 events (optionally plus or minus 2 or 3 events)) can be identified. Figures 18 to 21 A graphical display of passive and active sensed signals in the classification of moderate OSA based primarily on obstructive apnea events on a common time scale showing indications of classification events (e.g., obstructive apnea, mixed obstructive / central apnea, central apnea, and hypopnea) is shown (see figure below). Figures 22 to 25 A graphical display of passive and active sensed signals in a classification of severe OSA based primarily on obstructive apnea events on a common time scale showing indications of classification events (e.g., obstructive apnea, mixed obstructive / central apnea, central apnea, and hypopnea) is shown (see figure below). Fig.24 An exemplary benefit of the synergy of the passive and active sensing techniques described herein may be considered. When combined with the passively sensed data in the middle panel, the motion signal of the top active sensing panel may in isolation be misinterpreted as no effort at certain times (e.g., between amplitude changes), which may confuse the system to suggest a central apnea, when such a signal may be more accurately classified as reflecting an obstructive apnea event, as indicated in the lower panel.

[0600] How central apnea looks in the data:

[0601] The most prominent feature of a central apnea is the lack of respiratory effort during the apnea visible in the active signal. Many examples do not have snoring. However, features of gasping / loud breathing can be detected from the spectrogram. This can be supplemented by confirming gasps / loud breathing from the passive signal. Classification of central apneas can be achieved by utilizing the active signal and the lack of irregular snoring in the passive signal. In some minor cases, loud, regular snoring can be detected before the actual central apnea event (e.g., training cases can be provided to a deep neural network or a rule set can be used to allow classification of both cases). Figures 26 to 30 A graphical display of passive and active sensed signals in the classification of moderate OSA based primarily on central apnea events on a common time scale showing indications of classification events (e.g., obstructive apnea, mixed obstructive / central apnea, central apnea, and hypopnea) is shown (see figure below). Fig.31 A graphical display of passive and active sensed signals in the classification of severe OSA based primarily on central apnea events on a common time scale showing indications of classification events (e.g., obstructive apnea, mixed obstructive / central apnea, central apnea, and hypopnea) is shown (see figure below).

[0602] Central apnea in the data:

[0603] Hypopnea classification may be more difficult using only the passive signal. In this section, examples of hypopnea events from a severe subject are shown. For example, there is a visual example of a hypopnea sequence with no clear signature in the passive stream, while in another example, snoring / loud breathing is visible in the passive spectrogram (this particular subject showed signs of Cheyne-Stokes breathing and possible incipient snoring (34% of the total recording time according to the snoring detection algorithm). Figures 32 to 33 A graphical display of passive and active sensed signals in the classification of severe OSA based primarily on hypopnea events on a common time scale showing indications of classification events (e.g., obstructive apnea, mixed obstructive / central apnea, central apnea, and hypopnea) is shown (see figure below).

[0604] Snoring and obstructive apnea

[0605] The snoring detection algorithm can provide useful features for SDB detection, especially event classification. For example, there can be a clear correlation between total snoring time and AHI. Recordings with a predominantly obstructive component have a high proportion of total sleep time (TST) snoring. Most recordings with a predominantly central component have low levels of snoring, however there are exceptions. It may be the case that for these anomalies, the detected snoring is actually primary snoring. This can be further analyzed by the system by looking at a more granular level (i.e., correlating snoring markers with SDB events).

[0606] doctor

[0607] Doctors can gain insight into the health of patients with COPD or other chronic conditions and potentially connect to patients who need care but have not yet recognized that they have chronic disease symptoms. Thus, the processing device and its application (app) can be configured to generate / transmit cough-related information to a clinical system for remote monitoring of cough-related information.

[0608] Aged Care / Independent Living

[0609] As background, there is a connected care value proposition for this system in the elder care market, with the goal of reducing the care burden, particularly in independent living, by incorporating active sonar and / or passive acoustic technology in elder care digital assisted living tools. The system has broad potential - for example, those in independent living may also suffer from asthma, COPD, CHF, dementia, Alzheimer's disease, etc. In 2015, 1 in 8 people worldwide were over the age of 60. By 2030, 1 in 6 people will be 60 years old.

[0610] Current “independent living” solutions are typically a variety of disparate, expensive hardware with basic sensors (e.g., door sensors, water flow, bed occupancy, PIR motion, pressure mats, pill boxes, loud phones, wearable motion devices). Instead, a more complete solution can be provided by combining passive audio and active sonar into smart phones and / or smart speakers around the home. The active sonar and passive acoustic analytics overview in this article can fill this gap by enabling the following in living environments (e.g., homes, assisted living facilities, nursing homes / elderly care facilities, etc.):

[0611] Sports and Activity Tracking:

[0612] This can be done on a macro scale to detect any motion in the room with active sensing (e.g., using CW or FMCW sonar [processing any available range "bins" - e.g., all ranges] and / or by passively sensing / listening for the characteristic sounds of motion).

[0613] For the latter (passive only), a quiet environment with a limited number of ambient interferers is required (e.g. a desk or ceiling fan or open window with some external noise is suitable, but loud speech or a high-volume TV nearby (not uncommon for elderly people with hearing loss) makes purely passive acoustic analysis difficult. In contrast, ultrasonic FMCW or UWB acoustic systems can tolerate such interference better.

[0614] FMCW or UWB technology is also capable of localization range, so in addition to detecting any motion, the system is able to classify activity based on the intensity and duration of the motion and classify characteristic signatures of the activity.

[0615] Sit or lie down to breathe:

[0616] If their condition worsens (e.g. due to heart failure or difficulty breathing when lying down in people with COPD), the elderly may spend a lot of time sitting in front of the TV, and may actually spend more time there (i.e., trying to sleep in the chair).

[0617] This uses FMCW, (A)FHRG, UWB, etc. to locate possible breathing rates in space and track the user.

[0618] Sleep Stage Insights:

[0619] Sleep fragmentation measures are tracked, for example by detecting degradation in sleep measures as an indicator of worsening dementia and Alzheimer's disease.

[0620] Passive Hearing Disease Progress Insights:

[0621] This involves detecting changes in acoustic features over time, such as the prevalence and intensity of coughs, and periods of abnormal breathing rates.

[0622] This can be achieved by placing one or more smart speakers in a home environment (e.g., a commercial smart device such as an Amazon Echo Dot placed next to the bed, a Google Home Mini in the living room / TV room, a Google Mini in a nearby or kitchen area, etc.).

[0623] Adding acoustic features to daily life (user interaction with the environment):

[0624] This system can passively analyze the characteristic sounds of refrigerator doors, microwaves, ovens, front doors, etc. when they are opened and transmit these sounds to the monitoring station. Other characteristic sounds that indicate events include microwave timer "ding" or doorbells. Such user-environment interaction signatures, such as those related to appliance usage, can provide an indication of the activity of the monitored person, which can then be used as an indication of the user's condition / health, such as related to the norm or consistency / pattern of the monitored person's activities. The absence of such activity within an expected time period can be considered an indication of concern, or as a need for more direct contact to check on the monitored person (e.g., an elderly person).

[0625] Alternatively, a small ultrasonic "clicker" or the like could be mounted on these doors to create a characteristic sound (which may be audible or inaudible to a human) so that the system can better determine the exact event that occurred. In contrast to the sensors commonly used, this clicker does not require batteries or electronics as it is a simple mechanical component that "clicks" at a high frequency. This could be a two-part device that attaches to the door and frame, or a single inertial device that attaches to the door with adhesive and / or magnets. Thus, monitoring of many aspects of the home environment can be achieved.

[0626] An active version of an ultrasonic "clicker" may be used to provide further information. An active clicker sends an audible or inaudible data modulated CW sound to indicate an event, such as the status or status change of a window or door.

[0627] 8.3.4.1.3 Output module 8830

[0628] You can refer to Figures 14 to 1734 to 38 are exemplary graphical indicators to consider exemplary outputs that can be displayed on a display of a processing device such as a computer or mobile phone based on the above-mentioned evaluation and signal. In this regard, any signal / information generated by the evaluation module can be generated as an output of the output module 8830. For example, the determination of the SDB risk (or SDB probability) is output in a graphical risk-o-meter. The risk-o-meter may include a pointer (e.g., a needle) and a scale (e.g., a scale from 1 to 100). Other suitable forms and / or scales may also be provided. The scale may be circular or arc-shaped so as to allow the pointer to indicate a reading from a pivot point along the scale. In some versions, discrete portions (ranges) of the scale may reflect different colors to provide different features along the scale. For example, the scale may present a range for normal features (e.g., 0 to 15) and dangerous sleeper features (e.g., 15 to 100). Dangerous sleeper features may be formed by multiple sub-features, such as high and low. In this example, the high risk range of the risky sleeper characteristic can be coded as red, ranging from 30 to 100. The low range of the risky sleeper characteristic can be coded as orange, ranging from 15 to 30. Similarly, the normal sleep characteristic can be formed by multiple sub-characteristics, such as low and nominal. In this example, the low risk range of the normal sleeper characteristic can be coded as yellow, ranging from 5 to 15. The nominal range of the normal sleep characteristic can be coded as green, ranging from 0 to 5. Figures 14 to 17 and Fig.34 In the example of , the risk probability value determined in the assessment module 8820 is applied by the output module 8830 to generate an indicator having a pointer indicating the determined risk on a graphical scale. A text label may also optionally be provided for identifying a sleep characteristic from the scale range. In some versions, as shown, the graphical indicator may also include a text label having an estimated AHI count determined by the classifier of the assessment module 8820. Fig.34 In the exemplary risk meter of , the curve (eg, a circular curve segment) may be filled in proportion to the color indication of the risk level associated with the calculated probability.

[0629] In some versions, such a range or risk assessment (e.g., a count or risk probability) can be forwarded to a practitioner or other interested party, or applied to a respiratory therapy controller to adjust the therapy provided by the RPT device. For example, such a risk assessment indicative of SDB can be used as an input to a processor and / or therapy controller of the RPT, such as when the assessment is transmitted to the RPT device from an external processing device or when it is determined by a processor of the RPT device. The processor or controller can then change the therapy pressure, therapy flow rate, or other therapy parameter used to address the risk of SDB based on the value of the risk of SDB.

[0630] Alternatively, if Fig.17As shown, the indicator may present a message that data collection is insufficient, such as if the sleep period includes sensing signals from a period that is less than a threshold (eg, 1 hour). Fig.35 Another example of such a user interface display associated with snoring information data and a sleep score for a sleep session is shown.

[0631] 8.3.5 Potential Advantages of Some Technical Examples

[0632] Some versions of this technology may provide any of the following:

[0633] (1) Contactless SDB screening using RF hardware sensors and / or sonar sensing devices using applications (eg, software).

[0634] (2) It can detect sleep stages, including accurate estimation of AHI.

[0635] (3) sleep / wake correction can be provided, for example, for subjects with severe SDB to improve the accuracy of predicting AHI and sleep stages;

[0636] (4) Both active (sonar or RF) and passive (microphone) sensors / signatures can be implemented in a synergistic fusion of technologies to produce results even if the system is not optimally positioned in the most beneficial sensing direction;

[0637] (5) Features from both active and passive sensing signals can be evaluated to distinguish central from obstructive events and other types of events.

[0638] (6) Involving the evaluation of longer time scale periodic modulations in respiration / audio / motion (both intensity and frequency) rather than detecting individual events alone makes this system more robust to common interferers / confounding agents.

[0639] (7) Capture not only modulations in effort or audio, but also movements associated with SDB-related arousal / recovery.

[0640] (8) The system can be configured to focus on detecting the person closest to the bed in order to avoid sensing other people who are far away from the contactless sensing device.

[0641] Glossary

[0642] For the purposes of this technical disclosure, one or more of the following definitions may apply in certain forms of the technology. In other forms of the technology, alternative definitions may apply.

[0643] 8.3.6 General

[0644] Air: In some forms of the present technology, air may be considered to mean atmospheric air, and in other forms of the present technology, air may be considered to mean some other combination of breathable gases, such as atmospheric air enriched with oxygen.

[0645] Continuous Positive Airway Pressure (CPAP): CPAP therapy refers to the application of a supply of air to the entrance of the airway at a pressure that is continuously positive relative to atmosphere, and preferably approximately constant throughout the patient's breathing cycle. In some forms, the pressure at the entrance of the airway will be slightly higher during exhalation and slightly lower during inspiration. In some forms, the pressure will vary between different breathing cycles of the patient, for example, increasing in response to detecting an indication of partial upper airway obstruction, and decreasing in the absence of an indication of partial upper airway obstruction.

[0646] 8.3.7 Aspects of the respiratory cycle

[0647] Apnea: Apnea is considered to occur when flow drops below a predetermined threshold for a sustained period of time (e.g., 10 seconds). Obstructive apnea is considered to occur when some obstruction of the airway does not allow air to flow despite the patient's efforts. Central apnea is considered to occur when apnea is detected due to reduced or absent breathing effort despite an open airway. Mixed apnea is considered to occur when reduced or absent breathing effort occurs simultaneously with airway obstruction.

[0648] Respiratory rate: The patient's spontaneous breathing rate, usually measured in breaths per minute.

[0649] Duty cycle: the ratio of inspiratory time Ti to total breathing time Ttot.

[0650] Effort (Breathing): Preferably, breathing effort will be considered the work done by a spontaneous breather in an attempt to breathe.

[0651] Expiratory portion of the respiratory cycle: the period from the start of expiratory flow to the start of inspiratory flow.

[0652] Flow limitation: Preferably, flow limitation will be considered a state in a patient's breathing where an increase in the patient's effort does not result in a corresponding increase in flow. Where flow limitation occurs during the inspiratory portion of the breathing cycle, it may be described as inspiratory flow limitation. Where flow limitation occurs during the expiratory portion of the breathing cycle, it may be described as expiratory flow limitation.

[0653] Hypopnea: Preferably, hypopnea will be considered a decrease in flow, rather than a cessation of flow. In one form, hypopnea may be considered to occur when flow drops below a threshold for a sustained period of time. Central hypopnea is considered to occur when hypopnea is detected due to a decrease in respiratory effort.

[0654] Hyperpnea: Increased air flow to above normal levels.

[0655] Inspiratory portion of the breathing cycle: Preferably, the time period from the start of the inspiratory flow to the start of the expiratory flow is considered to be the inspiratory portion of the breathing cycle.

[0656]

[00136] Patency (airway): The degree to which an airway is open or the degree to which an airway is open. An open airway is open. Airway patency may be quantified, for example a value (1) being open and a value zero (0) being closed (obstructed).

[0657] Positive End Expiratory Pressure (PEEP): The pressure above atmospheric pressure that exists in the lungs at the end of exhalation.

[0658] Peak Flow (Q Peak): The maximum value of the flow during the inspiratory portion of the respiratory flow waveform.

[0659] Respiratory flow, airflow, patient airflow, respiratory airflow (Qr): These synonymous terms may be understood to refer to the RPT device's estimate of respiratory airflow, as opposed to "true respiratory flow" or "true respiratory airflow", which is the actual respiratory flow experienced by the patient, typically expressed in liters / minute.

[0660] Tidal volume (Vt): The volume of air inspired or exhaled during normal breathing when no extra effort is exerted.

[0661] (Inspiratory) Time (Ti): Duration of the inspiratory portion of the respiratory flow waveform.

[0662] (Expiratory) Time (Te): Duration of the expiratory portion of the respiratory flow waveform.

[0663] (Total) Time (Ttot): The total duration between the start of the inspiratory portion of one respiratory flow waveform and the start of the inspiratory portion of the subsequent respiratory flow waveform.

[0664] Typical Recent Ventilation: The ventilation value around which recent values ​​tend to cluster over some predetermined time scale, ie, a measure of the central tendency of recent values ​​of ventilation.

[0665] Upper airway obstruction (UAO): includes partial and total upper airway obstruction. This may be associated with a state of flow limitation, where the flow level increases only slightly, or even decreases, as the pressure difference across the upper airway increases (Starling impedance behavior).

[0666] Ventilation (Vent): A measure of the total amount of gas exchanged by a patient's respiratory system per unit time, including both inspiratory and expiratory flow. When expressed as volume per minute, this quantity is often referred to as "minute ventilation". Minute ventilation is sometimes given simply as volume, and is understood to be the volume per minute.

[0667] 8.3.8 RPT device parameters

[0668] Flow (or flow): The instantaneous volume (or mass) of air delivered per unit time. Although flow and ventilation have the same volume or mass scale per unit time, flow rate is measured over a much shorter period of time. In some cases, references to flow will be references to a scalar quantity, i.e., a quantity having only magnitude. In other cases, references to flow will be references to a vector quantity, i.e., a quantity having both magnitude and direction. When it is referred to as a signed quantity, flow can be nominally positive for the inspiratory portion of the patient's breathing cycle, and therefore negative for the expiratory portion of the patient's breathing cycle. Flow can be given by the symbol Q. The total flow Qt is the flow of air leaving the RPT device. The ventilation flow Qv is the flow of air leaving the vent to allow flushing of exhaled gases. The leakage flow Ql is the flow that is unintentionally leaked from the patient interface system. The respiratory flow Qr is the flow of air received into the patient's respiratory system.

[0669] Leakage: Preferably, the word leakage is the flow of air to the surrounding environment. Leaks may be intentional, for example to allow washing of exhaled CO2. Leaks may be unintentional, for example due to an incomplete seal between the mask and the patient's face. In one example, a leak may occur in the swivel elbow.

[0670] 8.4 Other notes

[0671] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office files or records, but otherwise reserves all copyright rights whatsoever.

[0672] Unless the context clearly indicates and provides a numerical range, it should be understood that each intervening value between the upper and lower limits of the range, to one-tenth of the unit of the lower limit, and any other such value or intervening value in the range are broadly included in the present technology. The upper and lower limits of these intermediate ranges may be independently included in the intermediate ranges and are also included in the present technology, but are subject to any explicitly excluded limits in the range. In the case where the range includes one or two of the limits, the present technology also includes ranges excluding either or both of those included limits.

[0673] Furthermore, where one or more values ​​are described herein for implementation as part of the present technology, it should be understood that such values ​​may be approximate and, unless otherwise indicated, may be used with any suitable significant digits as may be permitted or required by practical technical implementation.

[0674] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the technology belongs. Although any methods and materials similar or equivalent to those described herein can also be used in the practice or testing of the technology, a limited number of exemplary methods and materials are described herein.

[0675] When a particular material is considered to be preferred for use in constructing a component, obvious alternative materials having similar properties may be used as substitutes thereof. In addition, unless otherwise specified, any and all components described herein are understood to be capable of being manufactured and thus may be manufactured together or separately.

[0676] It must be noted that, as used herein and the appended claims, the singular forms "a," "an," and "the" include plural equivalents unless the context clearly dictates otherwise.

[0677] All publications mentioned herein are incorporated herein by reference to disclose and describe methods and / or materials as the subject of those publications. Publications discussed herein are provided only because they are disclosed before the filing date of the present application. Anything herein should not be construed as admitting that the present technology has no right to precede these publications due to previous inventions. In addition, the publication date provided may be different from the actual publication date, and the publication date may need to be confirmed independently.

[0678] Furthermore, in interpreting the present disclosure, all terms should be interpreted in the broadest reasonable manner consistent with the context. In particular, the terms "comprises" and "comprising" should be interpreted as referring to elements, components or steps in a non-exclusive manner, indicating that the mentioned elements, components or steps may exist or be utilized or combined with other elements, components or steps not explicitly mentioned.

[0679] The subject headings used in the detailed description are included only for the convenience of the reader's reference and should not be used to limit the inventive subject matter found in the entire disclosure or claims. The subject headings should not be used to interpret the scope of the claims or claim limitations.

[0680] Although the present technology has been described herein with reference to specific embodiments, it should be understood that these embodiments only illustrate the principles and applications of the present technology. In some cases, terms and symbols may imply specific details that are not required for practicing the present technology. For example, although the terms "first" and "second" may be used, unless otherwise specified, they are not intended to indicate any order, but may be used to distinguish different elements. In addition, although the process steps in the method may be described or illustrated in a certain order, this order is not necessary. Those skilled in the art will recognize that this order may be modified, and / or its various aspects may be performed simultaneously or even synchronously.

[0681] Therefore, it should be understood that various modifications may be made to the exemplary embodiments and other arrangements may be designed without departing from the spirit and scope of the present technology. For example, other versions of the present technology may also be understood by considering the following descriptive paragraphs and examples. To this end, the present technology may also involve:

[0682] Example 1. A method of one or more processors for monitoring a sleep disordered breathing state of a person, the method in the one or more processors comprising:

[0683] extracting one or more respiration signals from one or more motion signals, the one or more motion signals being generated by active contactless sensing;

[0684] extracting one or more energy band signals from a passive signal, the passive signal being generated by passive contactless sensing, the passive signal representing acoustic information detected by the acoustic sensor;

[0685] evaluating the one or more energy band signals and / or the one or more respiration signals to generate one or more intensity signals representative of the intensity of the sleep disordered breathing modulation; and

[0686] One or more features derived from the one or more intensity signals are classified to produce one or more sleep disordered breathing measures.

[0687] Example 2. The method according to Example 1 further includes

[0688] A sleep disordered breathing indication is generated based on the one or more sleep disordered breathing measures.

[0689] Example 3. The method of Example 2 further comprising displaying and / or forwarding a sleep disordered breathing indication on a display.

[0690] Example 4. The method of any one of Examples 1 to 3, further comprising displaying and / or forwarding one or more sleep disordered breathing measures on a display.

[0691] Example 5. A method according to any one of Examples 1 to 4, wherein the one or more breathing signals include a breathing effort signal.

[0692] Example 6. A method according to any one of Examples 1 to 5, wherein the one or more sleep disordered breathing measures include a probability of sleep disordered breathing.

[0693] Example 7. A method according to any one of Examples 1 to 6, wherein the classification includes identifying one of a positive and a negative that a number of sleep disordered breathing events during a sleep period exceeds a threshold, and wherein one or more sleep disordered breathing measures include a binary flag representing the result of the identification.

[0694] Example 8. The method of example 7, when dependent on example 6, wherein the binary flag indicates positive when the probability exceeds a threshold.

[0695] Example 9. The method of Example 7, wherein the sleep disordered breathing event includes at least one of an apnea event and a hypopnea event.

[0696] Example 10. The method of any one of Examples 1 to 9, wherein the one or more sleep disordered breathing measures include an apnea-hypopnea index representing an estimate of the total number of apnea events and hypopnea events.

[0697] Example 11. The method of any one of Examples 1 to 10, when dependent on Example 6, further comprising generating a sleep stage adjustment factor based on probability.

[0698] Example 12. The method of Example 11 further comprises adjusting the sleep stage time according to the adjustment factor.

[0699] Example 13. The method according to any one of Examples 1 to 12 further includes generating a cluster flag signal based on the one or more strength signals, the cluster flag signal representing a time sequence that identifies whether SDB modulation exists.

[0700] Example 14. The method of example 13, wherein the cluster flag signal is generated based on a comparison between a value of the one or more intensity signals and a threshold value.

[0701] Example 15. The method of example 14, wherein a flag of the cluster flag signal is set to true when a value of the one or more strength signals is greater than a first strength threshold.

[0702] Example 16. The method of example 15, wherein the cluster flag signal is further set based on an evaluation of a value of a filtered signal compared to a second strength threshold, the filtered signal being derived by filtering the one or more strength signals.

[0703] Example 17. A method according to any one of Examples 1 to 16, wherein the one or more characteristics include one or more total sleep time proportions with SDB clusters.

[0704] Example 18. A method according to any one of Examples 1 to 17, wherein one or more characteristics include peak intensity or peak average intensity.

[0705] Example 19. The method of any one of Examples 1 to 18, further comprising generating a sleep-wake correction mask signal based on a generated cluster flag signal characterizing the presence of a detected SDB cluster.

[0706] Example 20. The method according to Example 19 further comprises applying a sleep-wake correction mask signal to a sleep segmentation process, wherein the wake classification instances are corrected to sleep instances according to the sleep-wake correction mask signal during the sleep segmentation process.

[0707] Example 21. The method of any one of Examples 1 to 20, wherein the generated sleep disordered breathing indication comprises a graphical risk-o-meter displayed on a display device, the graphical risk-o-meter comprising a needle and a scale.

[0708] Example 22. A method according to Example 21, wherein the scale is represented using an indication of a discrete range of sleep disordered breathing risk.

[0709] Example 23. The method of any one of Examples 1 to 22, wherein evaluating one or more respiratory signals to generate one or more intensity signals representative of sleep disordered breathing modulation intensity comprises:

[0710] generating an envelope signal;

[0711] normalizing the envelope signal; and

[0712] Generates spectral characteristics from a normalized envelope signal.

[0713] Example 24. The method of Example 23, wherein the spectral characteristics include a peak frequency of a power spectral density operating within a sleep disordered breathing frequency range.

[0714] Example 25. A method according to any of Examples 23 and 24, wherein the spectral characteristics include an in-band metric comprising a ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep apnea frequency range from the power spectral density operation.

[0715] Example 26. A method according to any of Examples 23 and 24, wherein the spectral characteristics include an in-band metric comprising a ratio of: (a) the power in a narrow band near a peak frequency of a power spectral density operation, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in a narrow band near the peak frequency.

[0716] Example 27. A method according to any one of Examples 25 and 26, wherein the in-band metric is an average metric derived from in-band metric values ​​from the I channel motion signal and the Q channel motion signal.

[0717] Example 28. The method of any one of Examples 1 to 27, wherein evaluating the one or more energy band signals to generate one or more intensity signals representative of sleep disordered breathing modulation intensity comprises:

[0718] combining one or more energy band signals;

[0719] generating an envelope signal from the combined band signals;

[0720] filtering and normalizing an envelope signal from the combined band signals; and

[0721] A spectral characteristic is generated from the filtered and normalized envelope signal.

[0722] Example 29. A method according to Example 28, wherein the spectral characteristics from the filtered and normalized envelope signal include an in-band metric comprising the ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep apnea frequency range from the power spectral density operation.

[0723] Example 30. A method according to Example 28, wherein the spectral characteristics from the filtered and normalized envelope signal include an in-band metric comprising the ratio of: (a) the power in a narrow band near the peak frequency of a power spectral density operation, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in a narrow band near the peak frequency.

[0724] Example 31. A method according to any one of Examples 1 to 30, wherein extracting one or more breathing signals from one or more motion signals includes combining multiple motion signals, each of the multiple motion signals being a motion signal representing motion from a detection range that is different from a detection range of other motion signals in the multiple motion signals.

[0725] Example 32. The method of Example 31, wherein combining comprises calculating a weight for each of the plurality of motion signals according to the respiratory frequency from the power spectral density, and determining a weighted average of the absolute values ​​of the plurality of motion signals.

[0726] Example 33. The method of any one of Examples 1 to 32, wherein extracting one or more energy band signals from the passive signal comprises:

[0727] Separating the acoustic frequencies of the passive signal into band signals by calculating a transform of the passive signal;

[0728] Calculating the energy value of the band signal; and

[0729] Calculate the average of the calculated energy values ​​for each band signal.

[0730] Example 34. The method of any one of Examples 1 to 33, wherein active contactless sensing includes sonar sensing using a microphone and a speaker.

[0731] Example 35. The method of any one of Examples 1 to 33, wherein the active contactless sensing includes radar sensing using a radio frequency transmitter and receiver.

[0732] Example 36. The method of any one of Examples 1 to 35, wherein active contactless sensing comprises frequency modulated continuous wave (FMCW) sensing.

[0733] Example 37. A method according to any one of Examples 1 to 36, wherein the passive contactless sensing includes acoustic sensing of breathing-related sounds using a microphone.

[0734] Example 38. The method according to Example 37 also includes preprocessing the sound signal generated by the microphone to produce a passive signal, the preprocessing including one or more of the following: filtering using an infinite impulse response filter; removing the baseline, including subtracting the minimum value over a sliding window; removing artifacts, using percentile limits; normalizing with the standard deviation over a sliding window; and integration and high-pass filtering.

[0735] Example 39. The method according to Example 38 further includes performing autocorrelation processing on the pre-processed sound signal.

[0736] Example 40. The method of Example 39 further comprises detecting a peak of a predefined breathing range of the autocorrelated, preprocessed sound signal.

[0737] Example 41. The method according to Example 40 also includes determining a breathing rate estimate based on peak values ​​of multiple signals, each of the multiple signals is a sound signal in a discrete frequency band and is processed by preprocessing and autocorrelation.

[0738] Example 42. A method according to any one of Examples 1 to 41, wherein the one or more features are multiple features derived from multiple intensity signals from one or more energy band signals and one or more respiratory signals, and the generated one or more sleep breathing disorder measures are generated by classifying the multiple features.

[0739] Example 43. A method according to any one of Examples 1 to 42, wherein one or more processors are in a processing device, and the processing device includes any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

[0740] Example 44. The method of any of Examples 1 to 43 further comprises controlling, using one or more processors, changes to therapy settings of a respiratory therapy device based on one or more sleep disordered breathing measures.

[0741] Example 45. A processor-readable medium having processor-executable instructions stored thereon, which, when executed by one or more processors, cause the one or more processors to monitor a person's sleep breathing disorder state, and the processor-executable instructions are configured to execute a method as described in any one of Examples 1 to 44.

[0742] Example 46. A server capable of accessing the processor-readable medium of Example 45, wherein the server is configured to receive a request to download processor-executable instructions of the processor-readable medium to a processing device via a network.

[0743] Example 47. A processing device comprising: one or more processors; a speaker coupled to the one or more processors; a microphone coupled to the one or more processors; and a processor-readable medium as described in Example 45.

[0744] Example 48. A processing device according to Example 47, wherein the processing device comprises any one of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

[0745] Example 49. A processing device comprising: one or more processors; a microphone coupled to the one or more processors; a radio frequency sensor coupled to the one or more processors; and a processor-readable medium as described in Example 45.

[0746] Example 50. A method for a server capable of accessing a processor-readable medium as described in Example 45, the method comprising receiving at the server a request to download processor-executable instructions of the processor-readable medium to a processing device via a network; and transmitting the processor-executable instructions to the processing device in response to the request.

[0747] Example 51. A device for monitoring a person's sleep apnea state, the device comprising:

[0748] one or more sensors, the one or more sensors configured for active contactless sensing and passive contactless sensing; and

[0749] One or more processors, one or more processors configured as:

[0750] extracting one or more respiration signals from one or more motion signals, the one or more motion signals generated by active contactless sensing using one or more sensors;

[0751] extracting one or more energy band signals from a passive signal, the passive signal being generated by passive non-contact sensing using one or more sensors, the passive signal representing acoustic information detected by an acoustic sensor in the one or more sensors;

[0752] evaluating the one or more energy band signals and / or the one or more respiration signals to generate one or more intensity signals representative of the intensity of the sleep disordered breathing modulation; and

[0753] One or more features derived from the one or more intensity signals are classified to produce one or more sleep disordered breathing measures.

[0754] Example 52. The device of Example 51, wherein the one or more processors are further configured to generate a sleep disordered breathing indication based on the one or more sleep disordered breathing measures.

[0755] Example 53. A device according to Example 52, wherein the one or more processors are further configured to display and / or forward the sleep disordered breathing indication on a display.

[0756] Example 54. A device according to any of Examples 51 to 53, wherein the one or more processors are further configured to display and / or forward one or more sleep disordered breathing measures on a display.

[0757] Example 55. An apparatus according to any one of Examples 51 to 54, wherein the one or more breathing signals include a breathing effort signal.

[0758] Example 56. An apparatus according to any of Examples 51 to 55, wherein one or more sleep disordered breathing measures include a probability of sleep disordered breathing.

[0759] Example 57. An apparatus according to any one of Examples 51 to 56, wherein the classification includes identifying, either positively or negatively, that a number of sleep disordered breathing events during a sleep period exceeds a threshold, and wherein one or more sleep disordered breathing measures include a binary flag representing the result of the identification.

[0760] Example 58. An apparatus according to example 57, when dependent on example 56, wherein the binary flag indicates affirmation when the probability exceeds a threshold.

[0761] Example 59. An apparatus according to Example 57, wherein the sleep disordered breathing event includes at least one of an apnea event and a hypopnea event.

[0762] Example 60. An apparatus according to any of Examples 51 to 59, wherein one or more sleep disordered breathing measures include an apnea-hypopnea index representing an estimate of the total number of apnea events and hypopnea events.

[0763] Example 61. An apparatus according to any one of Examples 51 to 60, when dependent on Example 54, wherein the one or more processors are further configured to generate a sleep stage adjustment factor based on probability.

[0764] Example 62. A device according to Example 61, wherein the one or more processors are further configured to adjust the sleep stage time according to the adjustment factor.

[0765] Example 63. An apparatus according to any one of Examples 51 to 62, wherein the one or more processors are further configured to generate a cluster flag signal based on the one or more intensity signals, the cluster flag signal representing a time series that identifies whether SDB modulation exists.

[0766] Example 64. The apparatus of Example 63, wherein the cluster flag signal is generated based on a comparison between a value of one or more strength signals and a threshold value.

[0767] Example 65. The apparatus of Example 64, wherein a flag of the cluster flag signal is set to true when a value of the one or more strength signals is greater than a first strength threshold.

[0768] Example 66. The apparatus of Example 65, wherein the cluster flag signal is further set based on an evaluation of a value of a filtered signal compared to a second strength threshold, the filtered signal being derived by filtering the one or more strength signals.

[0769] Example 67. An apparatus according to any of Examples 51 to 66, wherein the one or more characteristics include one or more total sleep time proportions with SDB clusters.

[0770] Example 68. An apparatus according to any of Examples 51 to 67, wherein one or more characteristics include peak intensity or peak average intensity.

[0771] Example 69. The apparatus of any one of Examples 51 to 68, wherein the one or more processors are further configured to generate a sleep-wake correction mask signal based on a generated cluster flag signal characterizing the presence of a detected SDB cluster.

[0772] Example 70. A device according to Example 69, wherein the one or more processors are further configured to apply a sleep-wake correction mask signal to a sleep segmentation process, in which the wake classification instance is corrected to a sleep instance according to the sleep-wake correction mask signal.

[0773] Example 71. An apparatus according to any of Examples 51 to 70, wherein the generated sleep disordered breathing indication includes a graphical risk-o-meter displayed on a display device, the graphical risk-o-meter including a pointer and a scale.

[0774] Example 72. An apparatus according to Example 71, wherein the scale is represented using an indication of a discrete range of risk for sleep disordered breathing.

[0775] Example 73. The apparatus of any of Examples 51 to 72, wherein to evaluate the one or more breathing signals to generate one or more intensity signals representative of sleep disordered breathing modulation intensity, the one or more processors are configured to:

[0776] generating an envelope signal;

[0777] normalizing the envelope signal; and

[0778] Generates spectral characteristics from a normalized envelope signal.

[0779] Example 74. An apparatus according to Example 73, wherein the spectral characteristics include a peak frequency of a power spectral density operating within a sleep disordered breathing frequency range.

[0780] Example 75. An apparatus according to any of Examples 73 to 74, wherein the spectral characteristics include an in-band metric comprising a ratio of: (a) a peak frequency of a power spectral density operation, and (b) power in a sleep apnea frequency range from the power spectral density operation.

[0781] Example 76. An apparatus according to any of Examples 73 to 74, wherein the spectral characteristics include an in-band metric comprising a ratio of: (a) the power in a narrow band near a peak frequency of a...

Claims

1. A method of one or more processors for monitoring a sleep disordered breathing state of a person, the method in the one or more processors comprising: extracting one or more respiration signals from one or more motion signals generated by active contactless sensing; extracting one or more energy band signals from a passive signal generated by passive contactless sensing, the passive signal representing acoustic information detected by the acoustic sensor; evaluating the one or more energy band signals and / or the one or more respiration signals to generate one or more intensity signals representative of apnea sleep disordered breathing modulation intensity; as well as One or more features derived from the one or more intensity signals are classified or transmitted for classification to produce one or more sleep disordered breathing measures.

2. The method according to claim 1, further comprising one or more of the following: (a) generating a sleep disordered breathing indication based on the one or more sleep disordered breathing measurements; (b) displaying and / or forwarding the sleep disordered breathing indication on a display; and (c) displaying and / or forwarding the one or more sleep disordered breathing measurements on a display.

3. The method according to claim 1, wherein: The classification includes one of positive and negative identification that a number of sleep disordered breathing events exceeding a threshold value are present during the sleep period, and wherein the one or more sleep disordered breathing measures include a binary flag indicating a result of the identification.

4. The method according to claim 3, wherein: The one or more sleep disordered breathing measures include a probability of sleep disordered breathing, and wherein the binary flag indicates the affirmative when the probability exceeds a threshold.

5. The method according to any one of claims 1 to 4, wherein: The one or more sleep disordered breathing measures include an apnea-hypopnea index representing an estimate of the total number of apnea events and hypopnea events.

6. The method according to any one of claims 1 to 4, wherein: The one or more sleep disordered breathing measures include a probability of sleep disordered breathing, and the method further includes: (a) generating a sleep stage adjustment factor based on the probability, and (b) adjusting sleep stage time according to the sleep stage adjustment factor.

7. The method according to any one of claims 1 to 4, further comprising: A cluster flag signal is generated based on the one or more intensity signals, the cluster flag signal representing a time sequence identifying the presence or absence of SDB modulation.

8. The method according to claim 7, wherein: The cluster flag signal is generated based on a comparison between a value of the one or more strength signals and a threshold value.

9. The method according to claim 8, wherein: The flag of the cluster flag signal is set to true when the value of the one or more intensity signals is greater than a first intensity threshold, and wherein the cluster flag signal is further set based on an evaluation of the value of a filtered signal compared to a second intensity threshold, the filtered signal being derived by filtering the one or more intensity signals.

10. The method according to any one of claims 1 to 4, wherein: The one or more features include: (a) The total sleep time ratio of one or more SDB clusters; (b) peak intensity or peak average intensity; and / or (c) a plurality of features derived from a plurality of intensity signals from the one or more energy band signals.

11. The method according to any one of claims 1 to 4, further comprising: A sleep-wake correction mask signal is generated based on a generated cluster flag signal characterizing the presence of a detected SDB cluster, and the sleep-wake correction mask signal is applied to a sleep segmentation process, in which wake classification instances are corrected to sleep instances according to the sleep-wake correction mask signal.

12. The method according to any one of claims 2 to 4, wherein: The generated sleep disordered breathing indicator includes a graphical risk-o-meter displayed on a display device, the graphical risk-o-meter including a needle and a scale.

13. The method according to any one of claims 1 to 4, wherein: The evaluating the one or more respiratory signals to generate one or more intensity signals representative of apnea sleep disordered breathing modulation intensity and classifying one or more features derived from the one or more intensity signals comprises: generating an envelope signal; normalizing the envelope signal; and A spectral characteristic is generated from the normalized envelope signal.

14. The method according to claim 13, wherein: The spectral characteristics include any one or both of the following: (1) The peak frequency of the power spectral density operation within the sleep disordered breathing frequency range; (2) An in-band metric comprising the ratio of: (a) the power in a narrow band about a peak frequency of operation of the power spectral density, and (b) the difference between the total power of the spectrum of operation of the power spectral density and the power in the narrow band about the peak frequency.

15. The method according to any one of claims 1 to 4, wherein: The evaluating the one or more energy band signals to generate one or more intensity signals representing the intensity of apnea sleep disordered breathing modulation and classifying one or more features derived from the one or more intensity signals comprises: combining energy band signals from the one or more energy band signals; generating an envelope signal from the combined energy band signal; filtering and normalizing the envelope signal from the combined energy band signal; and A spectral characteristic is generated from the filtered and normalized envelope signal.

16. The method according to claim 15, wherein: Spectral characteristics from the filtered and normalized envelope signal include: An in-band metric comprising the ratio of: (a) the power in a narrow band about the peak frequency of operation of the power spectral density, and (b) the difference between the total power of the spectrum of operation of the power spectral density and the power in the narrow band about the peak frequency.

17. The method according to any one of claims 1 to 4, wherein: The extracting one or more breathing signals from the one or more motion signals includes combining a plurality of motion signals, each of the plurality of motion signals being a motion signal representing motion from a detection range that is different from a detection range of other motion signals in the plurality of motion signals.

18. The method according to any one of claims 1 to 4, wherein: The extracting one or more energy band signals from the passive signal comprises: separating the acoustic frequencies of the passive signal into band signals by calculating a transform of the passive signal; Calculating the energy value of the band signal; and Calculate the average of the calculated energy values ​​for each band signal.

19. The method according to any one of claims 1 to 4, wherein: (1) The active contactless sensing includes any of the following: (a) sonar sensing using a microphone and a speaker; (b) radar sensing using a radio frequency transmitter and a receiver; and (c) frequency modulated continuous wave (FMCW) sensing; and (2) passive contactless sensing includes acoustic sensing of breathing-related sounds using a microphone.

20. The method according to any one of claims 1 to 4, wherein: The one or more processors are in a processing device, which includes any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

21. The method according to any one of claims 1 to 4, further comprising: One or more processors are used to control changes to therapy settings of a respiratory therapy device based on the one or more sleep disordered breathing measures.

22. The method according to any one of claims 1 to 4, further comprising: receiving, in the one or more processors, the generated one or more sleep disordered breathing measures; as well as (a) displaying the received one or more sleep disordered breathing measures on a display; or (b) transmitting the received one or more sleep disordered breathing measures to a local processing device via a data communication transmission.

23. A processor-readable medium having processor-executable instructions stored thereon, wherein when the processor-executable instructions are executed by one or more processors, the one or more processors are enabled to monitor a person's sleep breathing disorder state, and the processor-executable instructions are configured to execute the method as described in any one of claims 1 to 4.

24. A server capable of accessing the processor-readable medium of claim 23, wherein: The server is configured to receive a request to download the processor-executable instructions of the processor-readable medium to a processing device over a network.

25. A processing device comprising: one or more processors; a speaker coupled to the one or more processors; a microphone coupled to the one or more processors; And the processor-readable medium of claim 23.

26. The processing device according to claim 25, wherein The processing device includes any of a smart phone, a tablet computer, a general computing device, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

27. A processing device comprising: one or more processors; a microphone coupled to the one or more processors; a radio frequency sensor coupled to the one or more processors; And the processor-readable medium of claim 23.

28. A method for a server capable of accessing a processor-readable medium as described in claim 23, the method comprising receiving at the server a request to download the processor-executable instructions of the processor-readable medium to a processing device via a network; and transmitting the processor-executable instructions to the processing device in response to the request.

29. A device for monitoring a person's sleep apnea state, the device comprising: One or more processors, the one or more processors configured to: extracting one or more respiration signals from one or more motion signals generated by active contactless sensing using one or more sensors; extracting one or more energy band signals from a passive signal generated by passive contactless sensing using the one or more sensors, the passive signal representing acoustic information detected by an acoustic sensor of the one or more sensors; evaluating the one or more energy band signals and / or the one or more respiration signals to generate one or more intensity signals representative of apnea sleep disordered breathing modulation intensity; as well as One or more features derived from the one or more intensity signals are classified or transmitted for classification to produce one or more sleep disordered breathing measures.

30. The device of claim 29, further comprising the one or more sensors configured for active contactless sensing and passive contactless sensing.

31. The apparatus of claim 29, wherein: The one or more processors are further configured to: (a) generating a sleep disordered breathing indication based on the one or more sleep disordered breathing measures; (b) displaying on a display and / or forwarding the sleep disordered breathing indication; and / or (c) displaying on a display and / or forwarding the one or more sleep disordered breathing measures.

32. The apparatus of claim 29, wherein: For classification, the one or more processors are configured to identify one of a positive and a negative that a number of sleep disordered breathing events during a sleep period exceeds a threshold, and wherein the one or more sleep disordered breathing measures include a binary flag indicating a result of the identification.

33. The apparatus of claim 32, wherein: The one or more sleep disordered breathing measures include a probability of sleep disordered breathing, wherein the binary flag indicates the affirmative when the probability exceeds a threshold.

34. Apparatus according to any one of claims 29 to 33, wherein: The one or more sleep disordered breathing measures include an apnea-hypopnea index representing an estimate of the total number of apnea events and hypopnea events.

35. Apparatus according to any one of claims 29 to 33, wherein: The one or more sleep disordered breathing metrics include a probability of sleep disordered breathing, and wherein the one or more processors are further configured to: (a) generating a sleep stage adjustment factor based on the probability; and (b) adjusting the sleep stage time according to the adjustment factor.

36. Apparatus according to any one of claims 31 to 33, wherein: The one or more processors are further configured to generate a cluster flag signal based on the one or more intensity signals, the cluster flag signal representing a time sequence identifying the presence or absence of SDB modulation.

37. The apparatus of claim 36, wherein: The cluster flag signal is generated based on a comparison between a value of the one or more strength signals and a threshold value.

38. The apparatus of claim 37, wherein: The flag of the cluster flag signal is set to true when the value of the one or more intensity signals is greater than a first intensity threshold, and wherein the cluster flag signal is further set based on an evaluation of the value of a filtered signal compared to a second intensity signal, the filtered signal being derived by filtering the one or more intensity signals.

39. Apparatus according to any one of claims 29 to 33, wherein: The one or more features include: (a) the proportion of total sleep time with one or more SDB clusters; (b) peak intensity or peak average intensity; and / or (c) a plurality of features derived from a plurality of intensity signals from the one or more energy band signals.

40. Apparatus according to any one of claims 29 to 33, wherein: The one or more processors are also configured to generate a sleep-wake correction mask signal based on a generated cluster flag signal characterizing the presence of a detected SDB cluster; and wherein the one or more processors are also configured to apply the sleep-wake correction mask signal to a sleep segmentation process, in which the wake-up classification instance is corrected to a sleep instance according to the sleep-wake correction mask signal.

41. Apparatus according to any one of claims 31 to 33, wherein: The generated sleep disordered breathing indicator includes a graphical risk-o-meter displayed on a display device, the graphical risk-o-meter including a needle and a scale.

42. Apparatus according to any one of claims 29 to 33, wherein: To evaluate the one or more respiratory signals to generate one or more intensity signals representative of apnea sleep disordered breathing modulation intensity and to classify one or more features derived from the one or more intensity signals, the one or more processors are configured to: generating an envelope signal; Normalizing the envelope signal; as well as A spectral characteristic is generated from the normalized envelope signal.

43. The apparatus of claim 42, wherein: The spectral characteristics include any one or both of the following: (1) The peak frequency of the power spectral density operation within the sleep disordered breathing frequency range; (2) An in-band metric comprising the ratio of: (a) the power in a narrow band about a peak frequency of operation of the power spectral density, and (b) the difference between the total power of the spectrum of operation of the power spectral density and the power in the narrow band about the peak frequency.

44. Apparatus according to any one of claims 29 to 33, wherein: To evaluate the one or more energy band signals to generate one or more intensity signals representative of apnea sleep disordered breathing modulation intensity and classify one or more features derived from the one or more intensity signals, the one or more processors are further configured to: combining energy band signals from the one or more energy band signals; generating an envelope signal from the combined energy band signal; as well as filtering and normalizing the envelope signal from the combined energy band signal; as well as A spectral characteristic is generated from the filtered and normalized envelope signal.

45. The apparatus of claim 44, wherein: Spectral characteristics from the filtered and normalized envelope signal include: An in-band metric comprises the ratio of: (a) the power in a narrow band about a peak frequency of operation of the power spectral density, and (b) the difference between the total power of the spectrum of the power spectral density operation and the power in the narrow band about the peak frequency.

46. ​​Apparatus according to any one of claims 29 to 33, wherein: In order to extract one or more breathing effort signals from one or more motion signals, the one or more processors are configured to combine multiple motion signals, each of the multiple motion signals being a motion signal representing motion from a detection range that is different from a detection range of other motion signals in the multiple motion signals.

47. Apparatus according to any one of claims 29 to 33, wherein: In order to extract the one or more energy band signals from the passive signal, the one or more processors are configured to: separating the acoustic frequencies of the passive signal into band signals by calculating a transform of the passive signal; Calculating the energy value of the band signal; and Calculate the average of the calculated energy values ​​for each band signal.

48. Apparatus according to any one of claims 29 to 33, wherein: (1) The active contactless sensing includes any of the following: (a) sonar sensing, wherein the one or more sensors include a microphone and a speaker; (b) radar sensing, wherein the one or more sensors include a radio frequency transmitter and a receiver; and (c) frequency modulated continuous wave (FMCW) sensing; and (2) the passive contactless sensing includes acoustic sensing of breathing-related sounds, wherein the one or more sensors include a microphone.

49. Apparatus according to any one of claims 29 to 33, wherein: The one or more processors are in a processing device, which includes any of a general computing device, a smart phone, a tablet computer, a smart speaker, a smart TV, a smart watch, and a respiratory therapy device.

50. Apparatus according to any one of claims 29 to 33, wherein: The one or more processors are configured to control changes to therapy settings of a respiratory therapy device based on the one or more sleep disordered breathing measures.

51. Apparatus according to any one of claims 29 to 33, wherein: The one or more processors are further configured to: receiving said generated one or more sleep disordered breathing measures; as well as (a) displaying the received one or more sleep disordered breathing measures on a display; or (b) transmitting the received one or more sleep disordered breathing measures to a local processing device via a data communication transmission.

Images