WO2024224391A1

WO2024224391A1 - Sleep apnea event determination using terahertz radar

Info

Publication number: WO2024224391A1
Application number: PCT/IL2024/050370
Authority: WO
Inventors: Dana Shavit
Original assignee: Neteera Technologies Ltd
Current assignee: Neteera Technologies Ltd
Priority date: 2023-04-24
Filing date: 2024-04-16
Publication date: 2024-10-31
Anticipated expiration: 2025-10-24
Also published as: IL302379B1; IL302379A; IL302379B2

Abstract

Method and system for monitoring sleep apnea. During a subject monitoring phase, a radar device receives a millimeter-wave reflection radar signal reflected from a monitored subject, the signal sampled and a signal portion at subject range including in-phase and quadrature components is extracted. A displacement signal reflecting micromovements associated with cardiac and pulmonary activity is derived and segmented into monitored subject segments, each having a selected segment duration corresponding to segment duration of reference subjects of a model training phase. An apnea event estimation model is applied to a dataset of monitoring samples, each sample including a respective monitored subject segment, to predict a number of apnea events of monitored subject segments over a monitored period. Sleep status of monitored subject is detected and overall sleep duration during monitored period is determined. Apnea-hypopnea index is determined based on predicted number of apnea events and determined sleep duration during monitored period.

Description

SLEEP APNEA EVENT DETERMINATION

USING TERAHERTZ RADAR

TECHNICAL FIELD

The present disclosure generally relates to the fields of high frequency radar sensors, temporal signal processing, machine learning, and sleep monitoring.

BACKGROUND

Sleep apnea is a type of sleeping disorder characterized by disruptions in breathing during sleep. An “apnea” more generally refers to a pause or temporal cessation of breathing, where a sleep apnea event is typically defined as a complete absence in the airflow of a sleeping subject for a minimum duration, such as for at least 10 seconds. Sleep apnea further encompasses “hypopneas” or periods of shallow or reduced breathing, where a hypopnea event is typically defined as a reduction in airflow of a sleeping subject by a minimum amount (e.g., at least 30%) for a minimum duration (e.g., at least 10 seconds). A primary classification of sleep apneas is “obstructive sleep apnea”, in which breathing is hindered by an obstruction in airflow, such as due to a narrowing or constriction in the upper airway. Another category is “central sleep apnea”, where repetitive or intermittent cessation of breathing results from instabilities in physiological mechanisms controlling respiration, such as neurological malfunctioning whereby the brain fails to signal activation of the diaphragm.

Apneas may occur several times over the course of a night. Symptoms may include loud snoring, gasping for air, frequent waking and poor-quality sleep during the night, followed by drowsiness and headaches in the daytime. Sleep apnea is a fairly common condition, with some estimates suggesting that it is experienced by nearly a quarter of the adult population globally. The propensity for sleep apnea is considered to increase significantly in older age groups.

Treatments for sleep apnea include the use of a continuous positive air pressure (CPAP) machine, as well as various lifestyle changes, such as losing weight, stopping smoking, and reducing alcohol consumption. A CPAP machine delivers a continuous stream of pressurized filtered air through a tube and into a mask worn over the nose and mouth of a sleeping person, keeping the airway open and preventing it from collapsing or narrowing. Other forms of treatment are oral appliances, such as a mandibular advancement device, which is designed to physically move the jaw and tongue forward so as to expand the airway. More acute treatment measures involve surgical procedures to enhance respiration.

Diagnosis of sleep apnea is typically implemented at a dedicated sleep testing facility via overnight monitoring and sleep testing known as polysomnography (PSG). The subject is connected to equipment that records changes in physiological parameters during sleep. For example, a polysomnogram may measure: brain activity (electroencephalography (EEG)), eye movements (electrooculography (EOG)), skeletal muscle activation (electromyography (EMG)), cardiac activity (electrocardiography (ECG)), oxygen saturation in the blood via pulse oximetry, as well as breathing function and respiratory effort, to evaluate for underlying causes of sleep disturbances. Such physiological measurements generally require cumbersome devices and sensors which need to be worn by or attached onto the body of the monitored person, and/or integrated into the recumbent surface, such as a bed, sofa, or mattress. Some measurements may utilize optical detection, such as using a visible-light camera or infrared (IR) sensor. However, such optical sensors require a direct line-of-sight to the body and clear visibility, and generally cannot function or provide degraded results under poor visibility conditions, or through obstructions or occlusions such as clothing or blanketing.

Radar based systems for sleep monitoring are known in the art. However, these systems generally require a calibration process for each room, surface and/or subject prior to use. Such calibrations can be exceedingly cumbersome and time-consuming, particularly when monitoring multiple subjects in a given room, on different surfaces, or in changing settings that may not necessarily be known in advance. Examples of publications directed to non-contact sleep monitoring and sleep apnea detection include the following:

U.S. patent application no. 2021/0177343 to Zhong et al, entitled: “Systems and methods for contactless sleep monitoring”; U.S. patent no. 10,624,574 to KONINKLIJKE PHILIPS NV, entitled: “Determination system and method for determining a sleep stage of a subject”;

China patent application No. CN112716474 to Fudan University, entitled: “Non-contact sleep state monitoring method and system based on biological microwave radar”;

Yang, Z., Pathak, P.H., Zeng, Y., Liran, X., & Mohapatra, P. (2017). Vital Sign and Sleep Monitoring Using Millimeter Wave. ACM Transactions on Sensor Networks (TOSN), 13, 1-32;

PCT patent application publication W02022/031038A1 to ASLEEP, entitled: “Computing device for predicting sleep state on basis of data measured in sleep environment of user”;

Anischchenko, Lesya, et al. “Sleep breathing disorders detection with bioradar using a long short-term memory network” In: 2020 XXXI I Ird General Assembly and Scientific Symposium of the International Union of Radio Science. IEEE, 2020, pp. 1-4; and

Taiwan patent application TW202239377A to Osense Technology Co Ltd., entitled: “Monitoring system and monitoring method for sleep apnea”.

SUMMARY

In accordance with one aspect of the present disclosure, there is thus provided a method for monitoring sleep apnea. During a model training phase, for each of a plurality of reference subjects, the method includes the steps of: receiving a THz or millimeter-wave reflection radar signal reflected from a respective reference subject; sampling the reflection radar signal and extracting a signal portion at a range of the reference subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; deriving a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the reference subject; segmenting the displacement signal into a plurality of reference subject segments, each of the reference subject segments having a selected segment duration; forming a training dataset comprising training samples obtained from a plurality of reference subjects, each training sample comprising a respective reference subject segment labeled with a measured number of apnea events during the segment duration; and applying at least one machine learning process to the training dataset to generate an apnea event estimation model. During a subject monitoring phase, the method includes the steps of: receiving a THz or millimeter-wave reflection radar signal reflected from at least one monitored subject; sampling the reflection radar and extracting a signal portion at a range of the monitored subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; deriving a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the monitored subject; segmenting the displacement signal into a plurality of monitored subject segments, each of the monitored subject segments having a selected segment duration corresponding to the segment duration of the reference subject segments; forming a monitoring dataset comprising monitoring samples obtained the monitored subject, each monitoring sample comprising a respective monitored subject segment; applying the apnea event estimation model to the monitoring samples to predict a number of apnea events of the monitored subject segments of the selected segment duration, over a monitored period; detecting a sleep status of the monitored subject and determining an overall sleep duration of the monitored subject during the monitored period; and determining an apnea-hypopnea index (AHI) of the monitored subject, based on the predicted number of apnea events and the determined sleep duration in the monitored time period. The radar signal may be obtained using a remote non-invasive radar device comprising: at least one radar transmitter, configured to transmit a radar signal to a body tissue of the subject; and at least one radar receiver, configured to receive a reflection of the transmitted radar signal reflected from the body tissue of the subject. The radar signal may be a frequency-modulated continuous-wave (FMCW) radar signal. Extracting a signal portion at a range of the subject may comprise applying a fast Fourier transform (FFT) to the FMCW radar signal. The signal portion may be sampled at a sampling rate of 500Hz. The method may further include the step of applying at least one processing operation to the displacement signal prior to the segmenting, during the model training phase or the subject monitoring phase, the processing operation selected from the group consisting of: bandpass filtering; normalization; and downsampling. The bandpass filtering may include filtering beyond a frequency range of 0.05Hz and 3.33Hz. The downsampling may include downsampling to a sampling rate of 10Hz. The selected segment duration may be 15 minutes. The model training phase may further include the step of establishing classification profiles of reference subjects by assigning reference subjects into different groups based on common features. The subject monitoring phase may further include detecting an occupancy of the monitored subject, and if the monitored subject is deemed absent in a selected period, updating the determined number of apnea events or the determined AHI accordingly. The method may include simultaneously monitoring multiple subjects in a location. The method may further include the step of determining AHI statistics of the monitored subject during subsequent monitoring sessions, the AHI statistics comprising at least one of: average AHI; peak AHI; and AHI standard deviation, over a plurality of sessions, and processing the AHI statistics to provide a focused behavioral recommendation for the monitored subject.

In accordance with another aspect of the present disclosure, there is thus provided a system for monitoring sleep apnea. The system includes a radar device and a processor. The radar device is configured to receive a THz or millimeter-wave reflection radar signal reflected from a reference subject during a model training phase, and to receive a THz or millimeter-wave reflection radar signal reflected from a monitored subject during a subject monitoring phase. During the model training phase, for each of a plurality of reference subjects, the processor is configured to sample the reflection radar signal, reflected from a respective reference subject, and to extract a signal portion at a range of the reference subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; to derive a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the reference subject; to segment the displacement signal into a plurality of reference subject segments, each of the reference subject segments having a selected segment duration; to form a training dataset comprising training samples obtained from a plurality of reference subjects, each training sample comprising a respective reference subject segment labeled with a measured number of apnea events during the segment duration; and to apply at least one machine learning process to the training dataset to generate an apnea event estimation model. During the subject monitoring phase, the processor is configured to sample the reflection radar signal, reflected from at least one monitored subject, and to extract a signal portion at a range of the monitored subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; to derive a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the monitored subject; to segment the displacement signal into a plurality of monitored subject segments, each of the monitored subject segments having a selected segment duration corresponding to the segment duration of the reference subject segments; to form a monitoring dataset comprising monitoring samples obtained from the monitored subject, each monitoring sample comprising a respective monitored subject segment; to apply the apnea event estimation model to the monitoring samples, to predict a number of apnea events of the monitored subject segments of the selected segment duration, over a monitored period; to detect a sleep status of the monitored subject and determine an overall sleep duration of the monitored subject during the monitored period; and to determine an apnea-hypopnea index (AHI) of the monitored subject, based on the predicted number of apnea events and the determined sleep duration in the monitored time period. The radar device may be a remote non-invasive radar device comprising: at least one radar transmitter, configured to transmit a radar signal to a body tissue of the subject; and at least one radar receiver, configured to receive a reflection of the transmitted radar signal reflected from the body tissue of the subject. The radar signal may be a frequency-modulated continuous-wave (FMCW) radar signal. Extracting a signal portion at a range of the subject may comprise applying a fast Fourier transform (FFT) to the FMCW radar signal. The signal portion may be sampled at a sampling rate of 500Hz. The processor may be further configured to apply at least one processing operation to the displacement signal prior to the segmenting, during the model training phase or the subject monitoring phase, the processing operation selected from the group consisting of: bandpass filtering; normalization; and downsampling. The bandpass filtering may include filtering beyond a frequency range of 0.05Hz and 3.33Hz. The downsampling may include downsampling to a sampling rate of 10Hz. The selected segment duration may be 15 minutes. During the model training phase, the processor may be further configured to establish classification profiles of reference subjects by assigning reference subjects into different groups based on common features. During the subject monitoring phase, the processor may be further configured to detect an occupancy of the monitored subject, and if the monitored subject is deemed absent in a selected period, to update the determined number of apnea events or the determined AHI accordingly. The system may include simultaneously monitoring multiple subjects in a location. The processor may be further configured to determine AHI statistics of the monitored subject during subsequent monitoring sessions, the AHI statistics comprising at least one of: average AHI; peak AHI; and AHI standard deviation, over a plurality of sessions, and to process the AHI statistics to provide a focused behavioral recommendation for the monitored subject.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure will be understood and appreciated more fully from the following detailed description taken in conjunction with the drawings in which: Figure 1 is a schematic illustration of a sleep apnea monitoring system, constructed and operative in accordance with an embodiment of the present disclosure;

Figure 2 is a schematic illustration of the system of Figure 1 applied on a subject lying on a bed, constructed and operative in accordance with an embodiment of the present disclosure;

Figure 3 is a block diagram of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure;

Figure 4 is a flow diagram of a data acquisition and model training phase of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure; and

Figure 5 is a flow diagram of a subject monitoring and posture classification phase of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure.

DETAILED DESCRIPTION OF THE EMBODIMENTS

The present disclosure may overcome the disadvantages of the prior art by providing a method and system for monitoring sleep apnea of a subject, with a high degree of accuracy and in a contact free manner, and without requiring a time-consuming calibration process for different environments, for different lying surfaces, and/or for different subjects.

Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the disclosed subject matter belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the specification and claims and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein. Well-known functions or constructions may not be described in detail for brevity and/or clarity.

It will be understood that, although the terms first, second, etc., may be used herein to describe various elements, components, regions, layers and/or sections, these elements, components, regions, layers and/or sections should not be limited by these terms. Rather, these terms are only used to distinguish one element, component, region, layer and/or section, from another element, component, region, layer and/or section.

It will be understood that when an element is referred to as being “on”, “attached” to, “operatively coupled” to, “operatively linked” to, “operatively engaged” with, “connected” to, “coupled” with, “contacting”, “added to, another element, it can be directly on, attached to, connected to, operatively coupled to, operatively engaged with, coupled with, added to, and/or contacting the other element or intervening elements can also be present. In contrast, when an element is referred to as being “directly contacting” another element or “directly added” to another element, there are no intervening elements and/or steps present.

Whenever the term “about” or “approximately” is used, it is meant to refer to a measurable value such as an amount, a temporal duration, and the like, and is meant to encompass variations (e.g., ±20%, ±10%, ±5%, ±1 %, ±0.1 %) from the specified value, as such variations are appropriate to perform the disclosed methods.

Certain features of the disclosure, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the disclosure, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable sub-combination or as suitable in any other described embodiment of the disclosure. Certain features described in the context of various embodiments are not to be considered essential features of those embodiments, unless the embodiment is inoperative without those elements.

Throughout this application, various embodiments of the present disclosure may be presented in a range format. It should be understood that the description in range format is merely for convenience and brevity and should not be construed as an inflexible limitation on the scope of the disclosed embodiments. Accordingly, the description of a range should be considered to have specifically disclosed all the possible subranges as well as individual numerical values within that range, regardless of the breadth of the range. For example, description of a range such as from 1 to 6 should be considered to have specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6 etc., as well as individual numbers within that range, for example, 1 , 2, 3, 4, 5, and 6. Whenever a numerical range is indicated herein, it is meant to include any cited numeral (fractional or integral) within the indicated range. For example, the phrases “ranging/ranges between” a first indicated number and a second indicated number and “ranging/ranges from” a first indicated number “to” a second indicated number are used herein interchangeably and are meant to include the first and second indicated numbers and all fractional and integral numerals there between.

Whenever terms “plurality” and “a plurality” are used it is meant to include, for example, “multiple” or “two or more”. The terms “plurality” or “a plurality” may be used throughout the specification to describe two or more components, devices, elements, units, parameters, or the like. The term set when used herein may include one or more items. Unless explicitly stated, the method embodiments described herein are not constrained to a particular order or sequence. Additionally, some of the described method embodiments or elements thereof can occur or be performed simultaneously, at the same point in time, or concurrently.

The term “repeatedly” as used herein should be broadly construed to include any one or more of: “continuously”, “periodic repetition” and “nonperiodic repetition”, where periodic repetition is characterized by constant length intervals between repetitions and non-periodic repetition is characterized by variable length intervals between repetitions.

The terms “user” and “operator” are used interchangeably herein to refer to any individual person or group of persons using or operating the method or system of the present disclosure, such as a person implementing a heartbeat interval measurement of a selected subject.

The term “subject” is used herein to refer to an individual upon which the method or system of the present disclosure is performed, such as a person for whom a sleep apnea/hypopnea index (AHI) is determined. The subject may be any living entity, such as a person, human or animal, characterized with a functioning heartbeat associated with a cardiac cycle of the heart.

The terms “sleep apnea event” and “apnea event”, are used interchangeably herein to refer to an instance of irregular breathing during sleep, including pauses in breathing (typically referred to as “apneas”), or periods of shallow or reduced breathing (typically referred to as “hypopneas”), where an apnea is defined as a complete absence in the airflow of a sleeping subject for a minimum duration (e.g., at least 10 seconds), and a hypopnea is defined as a reduction in airflow (e.g., by at least 30%) of a sleeping subject for a minimum duration (e.g., at least 10 seconds). A sleep apnea event may include obstructive apneas/hypopneas (i.e., caused by an obstruction or partial blockage in the subject airway), central apneas/hypopneas (i.e., caused by a reduced or failed attempt at breathing that may result from neurological malfunctioning), or a combination of both.

The term “Apnea-Hypopnea Index (AHI)” as used herein refers to a metric representing the average number of apnea and hypopnea events per hour of sleep, and is used as an indication of the severity of sleep apnea of a subject. Reference is now made to Figure 1 , which is a schematic illustration of a sleep apnea monitoring system, generally referenced 110, constructed and operative in accordance with an embodiment of the present disclosure. System 110 includes a radar device 112, a processor 114, a polysomnogram (PSG) device 116, and a database 118. Processor 114 is communicatively coupled with radar device 112, with PSG device 116, and with database 118.

Radar device 112 is configured to transmit a radar signal 122 to a body part of subject 120, such as the chest area, and to receive back a reflected radar signal 124. Radar device 112 includes at least one radar transmitter, and at least one radar receiver. Each of the radar transmitter and the radar receiver may include one or more transmitting/receiving elements, such as an individual radar antenna or an array of radar antennas, for example a phased array radar, such as a multiple-input multiple-output (MIMO) radar. The transmitted radar signal 122 is at a sufficiently high frequency to ensure that the signal is reflected and not absorbed by the body tissue, for example in the millimeter wave (MMW) frequency band (corresponding to EHF radio frequencies). According to an embodiment of the present disclosure, the transmitted and reflected radar signals 122, 124 may be in the Terahertz (THz) frequency band, where the term “Terahertz (THz)” as used herein encompasses Terahertz and sub-Terahertz radiation corresponding to sub-millimeter and millimeter wave radiation, such as electromagnetic waves within the frequency band between about 0.03 to 3 THz, corresponding to radiation wavelengths between about 10 mm to 0.1 mm.

Radar device 112 may be as described for example in PCT application publication WO2018/167777A1 to Neteera Technologies, entitled “Method and device for non-contact sensing of vital signs and diagnostic signals by electromagnetic waves in the sub terahertz band”, and PCT application publication W02020/012455A1 to Neteera Technologies, entitled “A sub-THz and THz system for physiological parameters detection and method thereof”. It is noted that radar device 112 operates in a contactless manner, which transmits and receives radar signals remotely without requiring a device component to be in direct physical contact with subject 120 or to be worn or attached to subject 120. It is further noted that radar device 112 may transit and/or receive a reflected radar signal from any direction of subject 120, such as from a front or back direction or from a non-orthogonal angle relative to subject 120. Moreover, radar device 112 may transmit and receive a reflected radar signal in low light or poor visibility conditions, as well as through certain obstructions or material barriers covering the subject body, where the radar signal may penetrate through clothing worn by subject 120, or a fabric or other material of a lying surface (e.g., bed, sofa or mattress) on which subject 120 is positioned.

Polysomnogram (PSG) device 116 records physiological information of a sleeping subject. PSG device 116 may obtain a polysomnogram or PSG recording, which may include readings relating to one or more physiological or bodily functions of the subject, such as: cardiac activity (e.g., ECG recording), brain activity (e.g., EEG recording), eye movements (e.g., EOG recording), and muscle activity (e.g., EMG recording), that can provide an indication of apnea events (i.e., apnea or hypopnea) during the monitored period. Accordingly, PSG device 116 may be embodied by one or more devices configured to detect particular physiological or bodily functions, including but not limited to: an instrument configured to detect cardiac activity, such as an electrocardiography (ECG) device; an instrument configured to detect brain activity, such as an electroencephalography (EEG) device; an instrument configured to detect eye movements or ocular activity, such as an electrooculography (EOG) device; and an instrument configured to detect muscular or skeletal muscle activity, such as an electromyography (EMG) device.

Processor 114 receives information or instructions from other components of system 110 and performs required data processing. For example, processor 114 receives and processes reflected radar signals 124 obtained by radar device 112 to generate a machine learning model during a data acquisition or modelling phase, and to determine a sleep apnea event index during a subject monitoring phase, as will be elaborated upon further hereinbelow.

Database 118 stores relevant information to be retrieved and processed by processor 114, such as radar signal data and associated information. Database 118 may be represented by one or more local servers or by remote and/or distributed servers, such as in a cloud storage platform.

Information may be conveyed between the components of system 110 over any suitable data communication channel or network, using any type of channel or network model and any data transmission protocol (e.g., wired, wireless, radio, WiFi, Bluetooth, and the like). For example, system 110 may store, manage and/or process data using a cloud computing model, and the components of system 110 may communicate with one another and be remotely monitored or controlled over the Internet, such as via an Internet of Things (loT) network. The components and devices of system 110 may be based in hardware, software, or combinations thereof. It is appreciated that the functionality associated with each of the devices or components of system 110 may be distributed among multiple devices or components, which may reside at a single location or at multiple locations. For example, the functionality associated with processor 114 may be distributed between a single processing unit or multiple processing units (e.g., a dedicated machine learning processor for the data modeling phase). Processor 114 may be part of a server or a remote computer system accessible over a communications medium or network, such as a cloud computing platform. Processor 114 may also be integrated with other components of system 110, such as incorporated with radar device 112.

System 110 may optionally include and/or be associated with additional components not shown in Figure 1 , for enabling the implementation of the disclosed subject matter. For example, system 110 may include a user interface (not shown) for allowing a user to control various parameters or settings associated with the components of system 110, a display device (not shown) for visually displaying information relating to the operation of system 110, and/or a camera or imaging device (not shown) for capturing images of the operation of system 110.

The operation of system 110 will now be described in general terms, followed by specific examples. Reference is made to Figure 2, which is a schematic illustration of the system 100 of Figure 1 applied on a subject 120 lying on a bed 130, constructed and operative in accordance with an embodiment of the present disclosure. System 110 is generally applied on a subject 120 situated in a lying or recumbent position, with the subject body aligned substantially horizontally parallel to the ground and supported by an underlying surface (e.g., a bed or sofa), such as while sleeping or resting. Radar device 112 is positioned in the vicinity of subject 120, such as mounted or held above the underlying surface. Radar device 112 is typically positioned a short distance away from subject 120, such as up to approximately 150 cm, but may generally be at further distances.

Reference is now made to Figure 3, which is a block diagram of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure. In procedure 152, a radar signal reflected from a subject is received. Referring to Figures 1 and 2, radar device 112 transmits a coherent radar signal 122, such as a frequency-modulated continuous wave (FMCW) radar signal in the THz frequency band, to a body part of subject 120, and receives a corresponding reflected radar signal 124, which contains information relating to micro displacements in the skin associated with the cardiac cycle and pulmonary activity of subject 120. Transmitted radar signal 122 may preferably be directed to a front body area of subject 120, such as the chest, but may also be directed to a rear body area, such as the back or neck.

In procedure 153, the reflected signal is sampled and a signal portion at the range of the subject is extracted. Referring to Figures 1 and 2, processor 114 receives and filters reflected radar signal 124 based on signal phase differences that correlate with distance, in order to isolate the signal components at the range at which subject 120 is located. In particular, processor 114 receives the reflected radar signal 124, obtained at a selected sampling rate and recorded with two channels consisting of an in-phase (I) component and a quadrature (Q) component, and measures a phase difference between the transmitted signal 122 and received signal 124 to determine the distance traversed by the radar signal, so as to extract the signal portion corresponding to reflections from subject 120 and removing noise and irrelevant signal components at other ranges. For example, if transmitted and reflected radar signals 122, 124 are FMCW radar signals, then processor 114 may apply a fast Fourier transform (FFT) to extract the signal portion corresponding to the subject range. For example, if operating in FMCW mode the reflected radar signal is sampled at a selected rate, e.g., 500 Hz, such that 500 times per second a vector of multiple samples (e.g., 128 samples) is collected (e.g., providing 64kHz samples per second). This received signal (e.g., of 128 samples) undergoes a Fourier transform, such that each transformed sample is respective of a range. Of these transformed samples, an individual sample corresponding to the subject range is extracted, resulting in a collection of values in accordance with the sampling rate (e.g., 500 extracted subject range samples per second). A determination of the subject range may be implemented, for example, in accordance with methods described in PCT application publication WO2013/275865A1 to Neteera Technologies, entitled “Radar-based range determination and validation”.

In procedure 154, a signal reflecting body displacement of the subject is derived. Referring to Figures 1 and 2, processor 114 receives reflected radar signal 124 and derives a body displacement signal, where the term “body displacement signal” or “displacement signal” as used herein is a signal reflective of ballistic forces associated with cardiac activity and pulmonary activity, characterizing repetitive body micromovements resulting from blood flow and ejection of blood into the vessels with each heartbeat as well as chest motion from inhalation and exhalation. In particular, processor 114 extracts the portion of reflected signal 124 corresponding to the range at which subject 120 is located, such as by measuring a phase difference between transmitted signal 122 and received signal 124 and applying a fast Fourier transform (FFT). The extracted signal portion (e.g., the output of the FFT, at the subject range, collected over predefined time intervals) then undergoes a non-linear filtering or mapping operation (e.g., involving the calculation of an angle of a complex number), to obtain a displacement signal.

In procedure 155, the displacement signal undergoes processing including bandpass filtering, normalization and down-sampling. Referring to Figures 1 and 2, processor 114 applies processing operations to the derived displacement signal, including bandpass filtering to remove very low frequencies and very high frequencies, for example ranging between 0.05Hz and 3.33Hz, such that frequencies pertaining to vital signs of the subject, such as respiration and heartrate and various harmonics, derivatives, traces and effects of physiological phenomena on the body, remain in the signal. Processor 114 may apply further processing operations, including normalization, and down-sampling to a selected sampling rate (for example, ranging from 500Hz to 10Hz). The processing steps may be interchangeable and reordered although bandpass filtering is usually implemented first. In procedure 156, the processed displacement signal is segmented. Referring to Figures 1 and 2, processor 114 divides the processed signal into segments of a selected time duration “N” (for example, approximately 15 minutes), to produce vectors of a selected length, where the segment duration is an adjustable parameter. For example, if the down-sampled frequency of the displacement signal is 10Hz, then a segmentation of 15 minutes provides a vector data set of length: 15 x 60 seconds x 10Hz = 9000.

The aforementioned process is repeated for multiple reference subjects to obtain a collection of training samples (vector segments) during an initial data acquisition and model training phase 160. In procedure 162 of the model training phase 160, training samples of reference vector segments are obtained and assigned with respective labels reflecting measured number of apnea events during the segment duration. Referring to Figure 1 , system 110 obtains a large number of vector segments from respective reference subjects 120, by performing steps 152, 153, 154, 155, 156, with each obtained vector segment representing a training sample. Each reference subject is monitored with polysomnogram device 116 from which sleep apnea information is derived. For each vector, processor 112 examines a PSG recording of the reference subject to compute an apnea-hypopnea index (AHI) reflecting the number of apnea events present during the time duration of that vector segment (e.g., 15 minutes). The PSG recording may include information or readings relating to one or more physiological or bodily functions of the subject, such as cardiac activity (e.g., ECG recording); brain activity (e.g., EEG recording), eye movements (e.g., EOG recording), and muscle activity (e.g., EMG recording), from which sleep apnea information may be extracted. Each training sample or vector is assigned a label of an apnea event count (e.g., 0, 1 , ... , N) reflecting the number of apnea events (apneas or hypopneas) of the reference subject detected during the duration of the vector segment. Each training sample dataset may optionally be assigned an additional subject identification label index, such as: a name, an identification number, and/or other personal information relating to the reference subject (e.g., age, gender, location, physical attributes) to facilitate subsequent data analysis. The vector datasets and associated apnea event labels are stored in database 118, and used as training data to generate a machine learning model that can be applied later for apnea event monitoring.

In procedure 164 of the model training phase 160, a machine learning process is applied to a collection of training samples for training an apnea event estimation model. The training dataset, representing vector segments and associated apnea event labels for a large number of reference subjects, is analyzed using a machine learning process, to implicitly identify different patterns and create models for estimating apnea events of monitored subjects. The machine learning process may apply machine learning techniques to analyze the training data, in order to produce mapping functions that can be used for classifying additional instances of new datasets (vector segments) according to relevant classification criteria. The data analysis may utilize any suitable machine learning or supervised learning process or algorithm, including but not limited to: an artificial neural network (ANN) process, such as a convolutional neural network, recurrent neural network (RNN), or a deep learning algorithm; a classification or regression analysis, such as a linear regression model; a logistic regression model, or a support-vector machine (SVM) model; a decision tree learning approach, such as a random forest classifier; and/or any combination thereof. The data analysis may utilize any suitable tool or platform, such as publicly available opensource machine learning or supervised learning tools.

Processor 114 may establish classification profiles of reference subjects by assigning reference subjects into different groups or categories based on common features. For example, the training process may provide models for identifying subjects in various categories, such as based on: age, gender, location, physical attributes, and the like. Each category may be associated with a relative weighting metric corresponding to a confidence level pertaining to the respective category feature. The different models may then be applied to facilitate the apnea event estimation of a new subject belonging to the relevant category or profile. The generated models may be iteratively updated and improved based on new information, such as accounting for subsequent successful or unsuccessful apnea event estimations of monitored subjects, and additional training data collected from new reference subjects. Simulations of numerous collected data and apnea event estimations may also be applied to enhance the reliability and accuracy of the models. The updated models may provide optimal formulas and weighting metrics for different variables or classification features. As more information and statistics are accumulated, the models can be further refined to improve their predictive capability for subject identification.

Reference is made to Figure 4, which is a flow diagram of a data acquisition and model training phase of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure. A reflection radar signal 212 is obtained from a reference subject 220 using a contactless radar device 112. Reflection radar signal 212 is sampled and filtered to extract a signal portion 214 at the subject range and to derive a displacement signal 216 representing physiological displacements or micro-movements of reference subject 220. Displacement signal 216 undergoes bandpass filtering, normalization, and down-sampling to produce a processed displacement signal 218, which is divided into segments of equal duration to form a reference data set 222 of vectors, each having a vector length corresponding to the segment duration. A PSG recording 224 is obtained from reference subject 220 using PSG device 116. Each vector in reference dataset 222 is assigned an apnea event label 226 based on information extracted from PSG recording 224, the apnea event label 226 reflecting the number of apnea events during the segment duration of the vector. This process is repeated for a large number of reference subjects 220 to generate a collection of training samples 228, each training sample made up of a vector segment 222 and its assigned apnea event label 226. The collection of training samples 228 is fed into a training model 230 that utilizes machine learning techniques to produce an apnea event estimation model 232 that can be applied on new input datasets. The generated (and dynamically updated) apnea event estimation model 232 is then used to estimate the number of apnea events and determine an apnea-hypopnea index (AHI) of a sleeping subject during a subject monitoring phase 170. It is appreciated that the training samples 228 and associated apnea event labels 226 may originate from other devices configured for measuring chest wall motion, in addition to or instead of a PSG device 116. Examples may include: a contactless radar device (e.g., radar device 112); accelerometers or motion sensors worn or coupled to a body part of the subject (such as the chest, back or abdomen); pressure sensors situated in or under the lying surface of the subject (e.g., bed or mattress); and the like. A collection of training samples 228 may include samples obtained from different sources, which may provide variation to enrich or augment the training process.

Reference is made to Figure 5, which is a flow diagram of a subject monitoring and posture classification phase of a sleep apnea monitoring method, operative in accordance with an embodiment of the present disclosure. A dataset of monitoring samples of a monitored subject 320 is obtained according to a similar sequence described hereinabove in the context of model training phase 160. In particular, a reflection radar signal 312 is obtained from a monitored subject 320 using a contactless radar device 112. Reflection radar signal 312 is sampled and filtered to extract a signal portion 314 at the subject range and to derive a displacement signal 316 representing physiological displacements or micro-movements of monitored subject 320. Displacement signal 316 undergoes bandpass filtering, normalization, and down-sampling to produce a processed displacement signal 318, which is divided into segments of equal duration to form a dataset of vector segments, each having a vector length corresponding to the segment duration. The vector segments, accumulated over the selected sample duration, make up monitoring samples 324, which are fed into apnea event estimation model 232.

Referring back to Figure 3, in procedure 172 of the subject monitoring phase 170, monitoring samples of segments from a monitored subject are obtained, each segment captured over a selected segment duration. Referring to Figure 5, monitoring dataset 326, made up of a plurality of samples 324 of vector segments derived from radar reflections 312 from monitored subject 320, are collected over a given time period and fed into apnea event estimation model 232. Vector segments of monitoring samples 324 are characterized by a vector length corresponding to the segment duration of the segments (applied when segmenting processed displacement signal 318), corresponding to the segment duration of the reference subject segments. Monitoring samples 324 may be provided to apnea event estimation model 232 as a continuous input stream, such that the displacement signals are segmented as acquired and the segments fed serially (one by one) into model 232. Alternatively, samples 324 may be delivered in parallel to apnea event estimation model 232 after waiting to acquire and segment a longer input signal. The particular segments of monitoring samples 324 that are processed may be dynamically selected. It is noted that consecutive segments may overlap, whereby consecutive segments of a certain duration (e.g., 10 second segments) may overlap by a certain amount (e.g., 9 seconds). Monitoring samples are collected over a selected monitoring period, such as over an entire night during when monitored subject 320 may be asleep.

In procedure 174 of the subject monitoring phase 170, an apnea event estimation model is applied to the monitoring samples to predict a number of apnea events of all segments of the selected duration over a monitored time period. Referring to Figure 5, apnea event estimation model 232 applies machine learning processes to the monitoring samples 324 to obtain an estimation of the number of apnea events present during the duration of the segments , based on earlier training of model 232 using labeled segments having the same duration. Model 232 provides an estimate of the number of apnea events of monitored subject 320 for each sample segment 324. Model 232 may further provide a prediction of the number of apnea events for all subsequent segments of the same duration (N) over a selected time period, such as during an entire interval when monitored subject 320 is asleep, based on monitoring samples 324 collected over the selected time period.

In procedure 176 of the subject monitoring phase 170, a sleep status of the monitored subject is detected, and an overall sleep duration of the monitored subject during the monitored time period is determined. Referring to Figures 1 and 5, processor 114 receives a sleep status indication 332 of monitored subject 320. Based on the sleep status indication 332, processor 114 determines the time period during which monitored subject 320 is actually asleep, and calculates an overall sleep duration (e.g., number of hours, minutes, and seconds) over the course of the monitoring session. Sleep status indication 332 may be obtained using PSG device 116 (e.g., derived from PSG recording 224) or from an alternate source, such as a separate machine learning process. Sleep status indication 332 may also be derived using supplemental information, such a signal reflecting the presence or absence of monitored subject 320 (e.g., obtained from an object detection sensor or proximity sensor), and/or a motion signal indicating movement of monitored subject 320 (e.g., obtained from a motion sensor). For example, if subject motion is detected and exceeds a certain threshold then it may indicate that the subject is in a wakened state. If a received signal denotes that the subject is absent (i.e. , not presently at the monitoring location), then no apnea event estimation is determined or voided during the absence, and an indication provided accordingly (e.g., a status of “absent” or “empty” is provided).

In procedure 178 of the subject monitoring phase 170, an apnea- hypopnea index (AHI) of the monitored subject is determined based on the number of predicted apnea events in the monitored time period divided by the sleep duration. Referring to Figures 1 and 5, processor 114 calculates an AHI value of monitored subject 320 based on the estimated number of apnea events obtained from model 232 (step 174) divided by the overall sleep duration or total number of sleep hours/minutes/seconds derived from sleep status signals 332 (step 176). The calculated AHI corresponds to the number of apnea and/or hypopnea events per hour of sleep, and represents a degree or severity of sleep apnea experienced by monitored subject 320. An exemplary categorization of sleep apnea severity for an adult may be as follows: “normal”: AHI<5; “mild sleep apnea”: 5<AHI<15; “moderate sleep apnea”: 15<AHI<30; “severe sleep apnea”: AHI>30. It is noted that if a given segment contains an amount of awake (nonsleep) time that exceeds a certain threshold (e.g., a high percentage of seconds in which the subject is in a wakened state), then that segment may be removed from the AHI derivation.

Monitored subject 320 may undergo additional monitoring in subsequent sessions, such as over a period of several days, weeks, or months, from which further sleep apnea event information and statistics may be determined (e.g., average AHI over multiple monitoring sessions; peak AHI; standard deviations; and the like). Such statistics may be reviewed and analyzed to provide focused behavioral recommendations for monitored subject 320, in order to alleviate or remedy sleep apnea and enhance quality of sleep. The method of Figure 3 is generally implemented in an iterative manner, such that at least some of the procedures are performed repeatedly, in order to provide for a dynamic determination of sleep apnea events and AHI (reflecting apnea seventy) of one or more subjects in real-time. It is appreciated that the disclosed embodiments may provide a reliable and accurate assessment of sleep apnea seventy, according to the determined AHI metric corresponding to the number of apnea and/or hypopnea events per actual hour of sleep time of the monitored subject over time, as determined using machine learning processes. Further parameters and calculated AHI statistics obtained over subsequent sessions can provide a more comprehensive evaluation and form the basis for providing targeted recommendations to alleviate the sleep apnea. The disclosed system and method does not require components to be in direct physical contact with the monitored subject, and there is no need for coupling a sensor or other device to the subject body before a monitoring session, thus saving time and minimizing discomfort. Furthermore, the subject does not need to be directly visible to the radar device, which may operate under poor visibility or light saturation conditions. The radar signal may be reliably obtained through obstructions or occlusions, such as clothing or blankets, and from different angles in relation to the subject. Source separation techniques may be utilized to enable measuring and identifying multiple subjects concurrently. The disclosed system does not require costly equipment and has relatively few components and is relatively straightforward to operate and maintain. The AHI determination of the disclosed method and system may be implemented in a wide variety of locations (e.g., rooms or facilities), and sleeping surfaces (e.g., different types and sizes of beds), and different types of subjects (e.g., regardless of age, height, weight, or other physical characteristics), without requiring a timeconsuming and cumbersome calibration process prior to each particular monitoring session depending on the type of location, sleeping surface, or subject. Moreover, the machine learning analysis provides reliable and accurate predictive models, which can be iteratively refined to improve the sleep apnea event and AHI determination of new subjects based on new information.

The disclosed system and method can be used for various applications, ranging from home healthcare to medical diagnosis. For example, the disclosed system and method may be applied for monitoring patients to evaluate and improve sleep quality in an eldercare facility.

While certain embodiments of the disclosed subject matter have been described, so as to enable one of skill in the art to practice the disclosed subject matter, the preceding description is intended to be exemplary only. It should not be used to limit the scope of the disclosed subject matter, which should be determined by reference to the following claims.

Claims

1 . A method for monitoring sleep apnea, the method comprising the steps of: during a model training phase: for each of a plurality of reference subjects, receiving a THz or millimeter-wave reflection radar signal reflected from a respective reference subject; sampling the reflection radar signal and extracting a signal portion at a range of the reference subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; deriving a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the reference subject; segmenting the displacement signal into a plurality of reference subject segments, each of the reference subject segments having a selected segment duration; forming a training dataset comprising training samples obtained from a plurality of reference subjects, each training sample comprising a respective reference subject segment labeled with a measured number of apnea events during the segment duration; and applying at least one machine learning process to the training dataset to generate an apnea event estimation model, and during a subject monitoring phase: receiving a THz or millimeter-wave reflection radar signal reflected from at least one monitored subject; sampling the reflection radar signal and extracting a signal portion at a range of the monitored subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; deriving a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the monitored subject; segmenting the displacement signal into a plurality of monitored subject segments, each of the monitored subject segments having a selected segment duration corresponding to the segment duration of the reference subject segments; forming a monitoring dataset comprising monitoring samples obtained from the monitored subject, each monitoring sample comprising a respective monitored subject segment; applying the apnea event estimation model to the monitoring samples, to predict a number of apnea events of the monitored subject segments of the selected segment duration, over a monitored period; detecting a sleep status of the monitored subject and determining an overall sleep duration of the monitored subject during the monitored period; and determining an apnea-hypopnea index (AHI) of the monitored subject, based on the predicted number of apnea events and the determined sleep duration in the monitored time period.

2. The method of claim 1 , wherein the radar signal is obtained using a remote non-invasive radar device comprising: at least one radar transmitter, configured to transmit a radar signal to a body tissue of the subject; and at least one radar receiver, configured to receive a reflection of the transmitted radar signal reflected from the body tissue of the subject.

3. The method of claim 1 , wherein the radar signal is a frequency-modulated continuous-wave (FMCW) radar signal.

4. The method of claim 3, wherein extracting a signal portion at a range of the subject comprises applying a fast Fourier transform (FFT) to the FMCW radar signal.

5. The method of claim 1 , wherein the signal portion is sampled at a sampling rate of 500Hz.

6. The method of claim 1 , further comprising the step of applying at least one processing operation to the displacement signal prior to the segmenting, during the model training phase or the subject monitoring phase, the processing operation selected from the group consisting of: bandpass filtering; normalization; and downsampling.

7. The method of claim 6, wherein the bandpass filtering comprises filtering beyond a frequency range of 0.05Hz and 3.33Hz.

8. The method of claim 6, wherein the downsampling comprises downsampling to a sampling rate of 10Hz.

9. The method of claim 1 , wherein the selected segment duration is 15 minutes.

10. The method of claim 1 , wherein the model training phase further comprises the step of establishing classification profiles of reference subjects by assigning reference subjects into different groups based on common features.

11. The method of claim 1 , wherein the subject monitoring phase further comprises detecting an occupancy of the monitored subject, and if the monitored subject is deemed absent in a selected period, updating the determined number of apnea events or the determined AHI accordingly.

12. The method of claim 1 , comprising simultaneously monitoring multiple subjects in a location.

13. The method of claim 1 , further comprising the step of determining AHI statistics of the monitored subject during subsequent monitoring sessions, the AHI statistics comprising at least one of: average AHI; peak AHI; and AHI standard deviation, over a plurality of sessions, and processing the AHI statistics to provide a focused behavioral recommendation for the monitored subject.

14. A system for monitoring sleep apnea, the system comprising: a radar device, configured to receive a THz or millimeter-wave reflection radar signal reflected from a reference subject during a model training phase, and to receive a THz or millimeter-wave reflection radar signal reflected from a monitored subject during a subject monitoring phase; and a processor, wherein during the model training phase, for each of a plurality of reference subjects, the processor is configured to sample the reflection radar signal, reflected from a respective reference subject, and to extract a signal portion at range of the reference subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; to derive a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the reference subject; to segment the displacement signal into a plurality of reference subject segments, each of the reference subject segments having a selected segment duration; to form a training dataset comprising training samples obtained from a plurality of reference subjects, each training sample comprising a respective reference subject segment labeled with a measured number of apnea events during the segment duration; and to apply at least one machine learning process to the training dataset to generate an apnea event estimation model, and wherein during the subject monitoring phase, the processor is configured to sample the reflection radar signal, reflected from at least one monitored subject, and to extract a signal portion at a range of the monitored subject, the signal portion consisting of an in-phase (I) component and a quadrature (Q) component; to derive a displacement signal reflecting body micromovements associated with cardiac and pulmonary activity of the monitored subject; to segment the displacement signal into a plurality of monitored subject segments, each of the monitored subject segments having a selected segment duration corresponding to the segment duration of the reference subject segments; to form a monitoring dataset comprising monitoring samples obtained from the monitored subject, each monitoring sample comprising a respective monitored subject segment; to apply the apnea event estimation model to the monitoring samples, to predict a number of apnea events of the monitored subject segments of the selected segment duration, over a monitored period; to detect a sleep status of the monitored subject and determine an overall sleep duration of the monitored subject during the monitored period; and to determine an apnea-hypopnea index (AHI) of the monitored subject, based on the predicted number of apnea events and the determined sleep duration in the monitored time period.

15. The system of claim 14, wherein the radar device is a remote non-invasive radar device comprising: at least one radar transmitter, configured to transmit a radar signal to a body tissue of the subject; and at least one radar receiver, configured to receive a reflection of the transmitted radar signal reflected from the body tissue of the subject.

16. The system of claim 1 , wherein the radar signal is a frequency-modulated continuous-wave (FMCW) radar signal.

17. The system of claim 16, wherein extracting a signal portion at a range of the subject comprises applying a fast Fourier transform (FFT) to the FMCW radar signal.

18. The system of claim 14, wherein the signal portion is sampled at a sampling rate of 500Hz.

19. The system of claim 14, wherein the processor is further configured to apply at least one processing operation to the displacement signal prior to the segmenting, during the model training phase or the subject monitoring phase, the processing operation selected from the group consisting of: bandpass filtering; normalization; and downsampling.

20. The system of claim 19, wherein the bandpass filtering comprises filtering beyond a frequency range of 0.05Hz and 3.33Hz.

21. The system of claim 19, wherein the downsampling comprises downsampling to a sampling rate of 10Hz.

22. The system of claim 14, wherein the selected segment duration is 15 minutes.

23. The system of claim 14, wherein during the model training phase, the processor is further configured to establish classification profiles of reference subjects by assigning reference subjects into different groups based on common features.

24. The system of claim 14, wherein during the subject monitoring phase, the processor is further configured to receive a detected occupancy of the monitored subject, and if the monitored subject is deemed absent in a selected period, to update the determined number of apnea events or the determined AHI accordingly.

25. The system of claim 14, comprising simultaneously monitoring multiple subjects in a location.

26. The system of claim 14, wherein the processor is further configured to determine AHI statistics of the monitored subject during subsequent monitoring sessions, the AHI statistics comprising at least one of: average AHI; peak AHI; and AHI standard deviation, over a plurality of sessions, and to process the AHI statistics to provide a focused behavioral recommendation for the monitored subject.