WO2023025770A1 - Sleep stage determining system - Google Patents
Sleep stage determining system Download PDFInfo
- Publication number
- WO2023025770A1 WO2023025770A1 PCT/EP2022/073417 EP2022073417W WO2023025770A1 WO 2023025770 A1 WO2023025770 A1 WO 2023025770A1 EP 2022073417 W EP2022073417 W EP 2022073417W WO 2023025770 A1 WO2023025770 A1 WO 2023025770A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- sleep stage
- cardiac activity
- sleep
- signals
- samples
- Prior art date
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
- A61B5/02405—Determining heart rate variability
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
- A61B5/02438—Detecting, measuring or recording pulse rate or heart rate with portable devices, e.g. worn by the patient
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
- G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1116—Determining posture transitions
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1118—Determining activity level
Definitions
- the present invention relates to a sleep stage determining system for determining a sleep stage and/or a sleep stage transition of a person.
- sleep abnormalities or pathologies are numerous and maybe as different as narcolepsy, sleepwalking, or abnormal sleep lengths such as insomnia or hypersomnia.
- sleep of a person maybe disturbed by snoring, which is often associated with the obstructive sleep apnoea syndrome, or by environmental factors, such as light or noise.
- sleep disturbances are generally marked by the occurrence of sleep events which combine sudden changes of physiological variables such as autonomic (respiratory or cardiac) or motor modifications.
- the sleep events can also be caused by symptoms of sleep pathologies such as sleep apnoea, restless leg, abnormal movement, sleepwalking, erratic heart rate, nightmare, night terror, etc.
- sleep pathologies such as sleep apnoea, restless leg, abnormal movement, sleepwalking, erratic heart rate, nightmare, night terror, etc.
- a snoring sleeper or talking and screaming during nightmare or night terror will usually cause an abnormal sleep event.
- the consequences of abnormal or disturbed sleep are numerous from a health (care) but also from a social-economic point of view.
- EEG electroencephalograms
- EOG electrooculogram
- EMG electromyogram
- REM Rapid Eye Movement
- a polysomnography provides various disadvantages. For instance, during an EEG electric potentials are recorded by using electrodes fixed on several sides of the skull, e.g. the electrodes are glued to the face and on the skull. Also, EOGs and EMGs require the attaching of electrodes and sensors to the face, skull or other body parts of the tested person. To detect eye movement during sleep, an EOG requires electrodes glued or otherwise attached near the eye or on the eyelid of the person.
- All these electrodes further require wires that are attached to the electrodes and lead to a device placed near the head of the bed limiting the movement freedom of the tested person. Such recording is therefore obtrusive due to wiring, unusual sleep environments and imposed schedule at bedding conditions. The results of such tests may therefore be distorted due to the changed environment of the tested person. Other devices are bulky and uncomfortable or require a data recorder.
- polysomnographies provide limitations due to the complexity of the recording techniques. In detail, specific recording places such as sleep laboratories and special equipment as well as well trained staff are necessary. Therefore, polysomnography remains an exceptional and expensive method for sleep evaluation.
- a polysomnography system based on detection of eyelid movement (EOG), head movement and a heart beat signal (electrocardiogram - ECG) is described in US 5,902,250.
- ECG eyelid movement
- head movement head movement
- a heart beat signal electrocardiogram - ECG
- US 5,902,250 is expensive and creates sleep disturbances due to the number of sensors and wires necessary.
- the system described in US 5,902,250 is not very precise in determining sleep stages and does not determine sleep stage transitions.
- WO 2012/156427 Al discloses a system and method for determining sleep, sleep stage and/or sleep stage transition of a person, including heart rate detecting means configured for detecting a heart rate of the person, movement detecting means configured for detecting a movement of a part of the body of the person, where the detected movement is caused by a skeletal muscle of the body, recording means configured for recording the detected heart rate and the detected movement of the part of the body, heart rate classifying means configured for classifying the recorded heart rate of the person into at least one heart rate class at least one heart rate variability class, movement classifying means configured for classifying the recorded movement into at least one movement class, and determining means configured for determining sleep, a sleep stage, a sleep stage transition and/or a sleep event of the person based at least partially on the at least one heart rate class and the at least one movement class.
- Two sensors of the same type may scatter and provide different signals even when measuring the same object or event. Further, a sensor used and sold with a respective signal receiving and processing device is different from the particular sensor used for developing a signal processing algorithm, here a sleep stage determining algorithm, run on the respective signal receiving and processing device. Further, the positioning of the sensor may affect its signals and thus the performance and accuracy of the sleep stage determining system.
- the present invention aims to provide an improved sleep stage determining system.
- the sleep stage determining system as described by claim 1 is suitable for determining a sleep stage and/or a sleep stage transition of a first person.
- the system comprises a sensor interface arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person’s cardiac activity.
- the system comprises a sleep stage model unit arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
- the acceleration sensor to be used by the first person may provide signals different from those of the acceleration sensor used to obtain the sleep stage model, particularly due to scatter.
- the system dispenses with an acceleration sensor. Thereby the problem is solved.
- the sleep stage model can be obtained by sensing cardiac activity signals of persons (probands) different from the first person, labelling different sleep stages in the obtained cardiac activity samples and subsequent machine learning using the labelled cardiac activity samples as training items.
- the sleep stage model can employ a similarity function k when processing the received cardiac activity signals.
- the sleep stage model can make use of pattern recognition or a support vector machine.
- the sleep stage model can be obtained by artificial intelligence including machine learning and/or rules from a sleep expert (expert rule).
- the cardiac activity signals can be classified by a sleep stage model which has been trained on multiple training items, preferably by machine learning.
- a training item includes the cardiac activity signals which have been classified into the respective categories or labelled, for example, by an expert.
- the labels associated with the cardiac activity signals can be considered to represent ground truth.
- a proband is a person who does not use the claimed system, in contrast to the first person (user).
- the proband’s cardiac activity is measured during a sampling interval and the obtained cardiac activity samples serve for training a sleep stage model to obtain a trained sleep stage model.
- the cardiac activity signal is an electrical heart signal, a pulse signal and/or a blood flow rate signal.
- the blood flow rate signal can be a blood volume flow rate signal.
- the cardiac activity signal maybe recorded using a singlelead electrocardiogram (ECG) or by monitoring the pulse rate through photoplethysmography (PPG) or ballistocardiography (BCG).
- ECG electrocardiogram
- PPG photoplethysmography
- BCG ballistocardiography
- Another embodiment further comprises a wearable device which includes the sensor interface, preferably also the sleep stage model unit.
- This embodiment may be more convenient during use and/or can be used outside a sleep laboratory.
- the sensor interface can be arranged in the wearable device together with a signal transmitting unit arranged for transmitting the cardiac activity signals to the sleep stage model unit located at a different location.
- the signal transmitting unit electrically connected to the sensor interface can be arranged to transmit the cardiac activity signals through a wire or wirelessly to the sleep stage model unit. This embodiment may be more convenient during use and/or can be used outside a sleep laboratory.
- a further embodiment comprises the cardiac activity sensor which is connected to the sensor interface.
- the cardiac activity sensor can be selected from a light sensor for photoplethysmography, a pressure sensor for ballistocardiography and an electrode for a single-lead electrocardiogram.
- the cardiac activity sensor can be integrated in the wearable device.
- the sleep stage model unit is arranged to receive as input only the cardiac activity signals and/or arranged to apply the sleep stage model only to the received cardiac activity signals, to determine a sleep stage and/or a sleep stage transition.
- the embodiment can dispense with an acceleration sensor.
- the sleep stage model unit of a preferred embodiment is further arranged to apply a heart rate model to the received cardiac activity signals to determine heart rate data (S11), a heart rate variability and/or a heart rate event.
- the heart rate model can be arranged to determine a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or two subsequent R-peaks of the QRS complex (S11’).
- the heart rate model can be arranged to determine whether and by how much the time interval varies over time.
- the heart rate model can be arranged to calculate a heart rate average, a variability value (including a temporal and spectral indices), a rhythm characteristic, a heart rate acceleration and/or a heart rate event or change from the received cardiac activity signals.
- the heart rate model can be configured for classifying a sleep stage and/or a sleep stage transition based on the calculated heart rate average, instantaneous heart rate values, variability value, rhythm characteristic and/or heart rate event or change.
- An embodiment comprises a sleep stage model unit which is arranged to receive as input the cardiac activity signals and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition, wherein the trained sleep stage model has been obtained by training a sleep stage model on multiple training items preferably by machine learning and/or artificial intelligence.
- the multiple training items comprise multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
- Supervised learning algorithms can be particularly suitable, such as classification, kernel machine, pattern recognition, regression, similarity learning, a support vector machine and transfer learning, among others. When two persons experience the same sleep stage, their cardiac activity signals may differ slightly.
- the trained sleep stage model may determine sleep stages and/or sleep stage transitions more accurately or reliably.
- the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing the cardiac activity of several probands, and by labelling different sleep stages in the obtained cardiac activity samples.
- Each of the multiple training items includes the obtained cardiac activity samples and the labelled sleep stages.
- the trained sleep stage model may determine sleep stages and/or sleep stage transitions more accurately or reliably.
- the sleep stage model unit of another embodiment is arranged to receive as input the cardiac activity signals, and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
- the trained sleep stage model has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples and multiple body motility samples of probands, and labelled sleep stages and/or sleep stage transitions.
- the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing the cardiac activity of several probands, by measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the probands, and by labelling different sleep stages in the obtained cardiac activity samples, such that the multiple training items include the obtained cardiac activity samples, the obtained body motility samples, and the labelled different sleep stages.
- An embodiment further includes a sampling device suitable for obtaining multiple training items during the sampling interval, the sampling device comprising a cardiac activity sensor for sensing a proband’s cardiac activity signals.
- the cardiac activity sensor of the sampling device is of the same type as the cardiac activity sensor of the above embodiments.
- the sampling device can further comprise a body motility sensor for sensing the proband’s body motility.
- the body motility sensor can be an acceleration sensor, a pressure sensor or a camera.
- a method of training a sleep stage model to be used with any of the above systems forms a second aspect of the invention.
- the method of training a sleep stage model comprises the steps:
- Step Si can further include measuring body motility.
- the labelling during step S2 can be based on polysomnography, either by “traditional manual” scoring by a sleep expert or a trained physician, or by automatic scoring using a validated algorithm.
- the labelled sleep stages and/or sleep stage transitions of the multiple training items can be considered to represent ground truth.
- This method may help to improve the accuracy or reliability of determining sleep stages and/or sleep stage transitions.
- a preferred method of training a sleep stage model further comprises:
- the multiple training items comprise the multiple cardiac activity samples and multiple body motility samples of probands and associated sleep stages and/or sleep stage transitions.
- a method of determining a sleep stage and/or a sleep stage transition of the first person employs one of the systems explained above. The method comprises the steps:
- the sleep stage model of step S21 is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
- a preferred method of determining a sleep stage and/or a sleep stage transition of the first person comprises above step Sio, obtaining heart rate data from the cardiac activity signals (Sil), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (Sil’), and applying the sleep stage model to the heart rate data to determine a sleep stage and/or a sleep stage transition (S21’).
- the sleep stage model of step S21’ is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples or heart rate data of probands and associated sleep stages and/or sleep stage transitions.
- Cardiac activity signals are obtained from the cardiac activity sensor sensing the first person’s cardiac activity signals (Sio).
- Heart rate data is obtained from the cardiac activity signals (S11), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (S11’).
- the heart rate data can be prefiltered and abnormalities detected (S12).
- the prefiltered heart rate data can be parameterised by filters, FFT or the like (S13).
- the parameterised signals can be used with a support vector machine (S14) and sleep stage probabilities can be obtained.
- the prefiltered heart rate data or heart rate signals can be examined for detecting physiological events (S15), such as cardiac arousals (AC), small-AC-like-Patterns (Salp) and variations in heart rate.
- Step S15 may be done through the use of pattern recognition and signal processing techniques, among others.
- the detected physiological events can be used for estimating body motility (S16), as movements have direct effects on the cardiac activity.
- the sleep stage probabilities, the detected physiological events and estimated body motility can contribute to the sleep stage model (S17).
- the sleep stage model includes sleep stage transition rules and permits to determine sleep stage transitions by these transition rules.
- the sleep stage model can include a) interpreting an increased estimated body motility and increased cardiac activity events as a sleep stage transition towards wakefulness, b) interpreting a reduced estimated body motility and reduced cardiac activity events as a sleep stage transition towards deep Non-REM sleep.
- the sleep stage model can determine a subsequent sleep stage from the present sleep stage if the right conditions are fulfilled among others, based on probability.
- the sleep stages can be classified with the sleep stage transitions (S18) and can be shown in a hypnogram. Steps S12 to S18 can be considered as details of step S21.
- a sleep stage classification can be performed by applying the above sleep stage transition rules.
- a first sleep state which can be a wake state at the beginning of the night
- the sleep stage model applies the sleep stage transition rules on the received cardiac activity signals for obtaining the most probable sleep stage transition (S25).
- the sleep stage transition rules permit the sleep stage model unit to compute the time of each possible transition applicable from the current sleep state, based on detected physiological events, pre-detected sleep events and previously determined sleep stages (S25). From the first sleep stage the most probable sleep stage transition leads to a subsequent sleep stage (S26). The next possible and most probable sleep stage transition at point in time Tn indicates the next sleep state (S26).
- Steps S25, S26 can be repeated along the received cardiac activity signals until the end of the night (“lights on”) is reached or no more sleep stage transitions are found.
- heart rate data can be obtained from the cardiac activity signals (S11), such that S25, S26 are performed on heart rate data.
- the sleep stages can be categorized as rapid eye movement (REM) sleep stage and non-REM sleep stages.
- the REM sleep stage is the one where vivid dreaming occurs. It can be identified by the occurrence of rapid eye movements under closed eyelids, motor atonia and low voltage EEG patterns.
- the REM sleep stage also referred to as REM sleep, is also associated with bursts of muscular twitching, irregular breathing, irregular heart rate and increased autonomic activity. Periods of REM sleep are also referred to as paradoxical sleep.
- the sleep of a person can also be scored into non-REM (NREM) stages, which are numbered 1 to 3.
- Figure 1 depicts an exemplary hypnogram of a healthy young adult showing the different sleep stages of an eight hour sleep recording. It must be noted that the transitions from one stage to another are conventionally considered as abrupt steps. As illustrated, within the first hour of sleep a person starting from a wakeful state and falling asleep may transit to NREM sleep stage 1 and further to stages 2, and 3.
- the criteria of sleep stage 1 of the NREM sleep consist of a low voltage EEG tracing with well defined alpha activity and theta frequencies in the 3 to 7 Hz range, occasional vertex spikes, and slow rolling eye movements (SEMs). This stage includes the absence of sleep spindles, K- complexes and REMs. Stage 1 normally represents 4 to 5% of the total amount of sleep time.
- the sleep stage 2 of NREM sleep is characterized by the occurrence of sleep spindles and K- complexes against a relatively low voltage, mixed frequency EEG background.
- High voltage delta waves may comprise up to 20% of stage 2 epochs.
- the sleep stage 2 usually accounts for 45 to 55% of the total sleep time.
- a light Non-REM sleep stage is a common term for the sleep stages 1 and 2 (light sleep), while a deep Non-REM sleep is a term for sleep stage 3.
- the rest of the sleep as depicted in Figure 1 comprises transitions from REM sleep periods to lighter Non-REM sleep stages, such as stages 1 and 2.
- the present invention provides a system for sensing and recording continuously and up to several days or weeks the basic physiological variables such as heart rate and body motility together with some characteristics of the ambient physical environment.
- This methodology will be able to score the basic states such as waking and sleeping periods of the tested person.
- sleep stages will be scored every 30-second epoch.
- simultaneous recording of ambient physical variables together with the biological ones will allow evaluating the possible impact of the former to the latter.
- Figure 2 shows an exemplary sleep stage determining system too for determining a sleep stage and/or a sleep stage transition of a first person.
- the system comprises a sensor interface 101 arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person’s cardiac activity, and a sleep stage model unit 102 arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
- the sensor interface and the sleep stage model unit can be arranged in a wearable device 106 (dashed line).
- the exemplary system can comprise a cardiac activity sensor 105 (dashed line) electrically connected to the sensor interface 101.
- the cardiac activity sensor can form a part of the wearable device.
- the sleep stage model unit can be at a different location.
- the sensor interface of another exemplary sleep stage determining system (not shown) is connected with a cardiac activity sensor but with no other sensor.
- a sleep stage model can be obtained by an exemplary method comprising sensing cardiac activity signals of probands during a sampling interval (Si) and labelling different sleep stages in the measured cardiac activity signals or cardiac activity samples (S2).
- the sleep stage model can employ a similarity function k when processing the received cardiac activity signals particularly of probands.
- Obtaining the sleep stage model can involve pattern recognition applied to the cardiac activity signals of several probands.
- a sleep stage model can be obtained by another exemplary method comprising steps Si, S2 and by training a sleep stage model on multiple training items , preferably by machine learning (S3), see figure 3, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
- S3 machine learning
- a sleep stage model can be obtained by a further exemplary method comprising steps Si, S2 and by training a sleep stage model on multiple training items, preferably by machine learning (S3), the multiple training items comprising multiple cardiac activity samples and multiple body motility samples of probands, and labelled sleep stages and/or sleep stage transitions.
- the body motility samples of probands are obtained by measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the probands (S4), see figure 3.
- An exemplary method of determining a sleep stage and/or a sleep stage transition of the first person employs one of the systems explained above.
- the exemplary method is shown in figure 4a and comprises the steps: S10 obtaining cardiac activity signals from the cardiac activity sensor sensing the first person’s cardiac activity signals,
- the sleep stage model of step S21 is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
- Cardiac activity signals are obtained from the cardiac activity sensor sensing the first person’s cardiac activity signals (S10).
- Heart rate data is obtained from the cardiac activity signals (S11), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (S11’).
- the heart rate data can be prefiltered and abnormalities detected (S12).
- the prefiltered heart rate data can be parameterised by filters, FFT or the like (S13).
- the parameterised signals can be used with a support vector machine (S14) and sleep stage probabilities can be obtained.
- the prefiltered heart rate data or heart rate signals can be examined for detecting physiological events (S15), such as cardiac arousals (AC), small-AC-like-Patterns (Salp) and variations in heart rate.
- Step S15 maybe done through the use of pattern recognition and signal processing techniques, among others.
- the detected physiological events can be used for estimating body motility (S16), as movements have direct effects on the cardiac activity.
- the sleep stage probabilities, the detected physiological events and estimated body motility can contribute to the sleep stage model (S17).
- the sleep stage model includes sleep stage transition rules and permits to determine sleep stage transitions by these transition rules.
- the sleep stage model can include a) interpreting an increased estimated body motility and increased cardiac activity events as a sleep stage transition towards wakefulness, b) interpreting a reduced estimated body motility and reduced cardiac activity events as a sleep stage transition towards deep Non-REM sleep.
- the sleep stage model can determine a subsequent sleep stage from the present sleep stage if the right conditions are fulfilled among others, based on probability.
- the sleep stages can be classified with the sleep stage transitions (S18), and can be shown in a hypnogram. The exemplary method is shown in figure 4b.
- a sleep stage classification can be performed by applying the above sleep stage transition rules.
- the sleep stage model applies the sleep stage transition rules on the received cardiac activity signals for obtaining the most probable sleep stage transition (S25).
- the sleep stage transition rules permit the sleep stage model unit to compute the time of each possible transition applicable from the current sleep state, based on detected physiological events, pre-detected sleep events and previously determined sleep stages (S25). From the first sleep stage the most probable sleep stage transition leads to a subsequent sleep stage (S26), see figure 6. The next possible and most probable sleep stage transition at point in time Tn indicates the next sleep state (S26), see figure 5.
- Steps S25, S26 can be repeated along the received cardiac activity signals until the end of the night (“lights on”) is reached or no more sleep stage transitions are found, see figures 5, 6.
- heart rate data can be obtained from the cardiac activity signals (S11), such that S25, S26 are performed on heart rate data, see figure 6.
Abstract
A sleep stage determining system for determining a sleep stage and/or a sleep stage transition of a first person, wherein the system comprises: a sensor interface arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person's cardiac activity, and a sleep stage model unit arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
Description
Sleep Stage Determining System
Description
[0001] The present invention relates to a sleep stage determining system for determining a sleep stage and/or a sleep stage transition of a person.
Background
[0002] A plurality of people faces problems due to sleep abnormalities and sleep disturbances. For example, sleep abnormalities or pathologies are numerous and maybe as different as narcolepsy, sleepwalking, or abnormal sleep lengths such as insomnia or hypersomnia. Moreover, the sleep of a person maybe disturbed by snoring, which is often associated with the obstructive sleep apnoea syndrome, or by environmental factors, such as light or noise. These sleep disturbances are generally marked by the occurrence of sleep events which combine sudden changes of physiological variables such as autonomic (respiratory or cardiac) or motor modifications. The sleep events can also be caused by symptoms of sleep pathologies such as sleep apnoea, restless leg, abnormal movement, sleepwalking, erratic heart rate, nightmare, night terror, etc. Thus, a snoring sleeper or talking and screaming during nightmare or night terror will usually cause an abnormal sleep event. The consequences of abnormal or disturbed sleep are numerous from a health (care) but also from a social-economic point of view.
[0003] In order to detect the reasons of the peoples’ sleep abnormalities and disturbances, sleep laboratories can conduct a sleep scoring of the person, i.e. the determination of sleep stages and their transitions. In a sleep laboratory, physiological parameters are observed and corresponding data recorded in a polysomnography. This recording during a polysomnography includes primary data such as electroencephalograms (EEG), electrooculogram (EOG) and electromyogram (EMG), and secondary data such as heart rate, respiration, oximetry and body movements. EEG is used to detect and name brainwaves according to their frequency and amplitude. With an EOG
the movement of the eye balls is recognized and analysed. EMG allows for evaluating and recording the electrical activity produced by skeletal muscles.
[0004] Classically, sleep scoring is based on the analysis of EEG, EOG and EMG recordings made continuously throughout the sleep period. These physiological data are represented by fluctuations of electrical potentials recorded by small electrodes attached to different parts of the scalp and the face of the tested/recorded person.
[0005] Those electrical potentials are then interpreted by a sleep specialist according to internationally accepted rules which define the different stages of sleep. Each sleep stage is characterized by the presence and the abundance of specific EEG waves on the recording. Further, eye movements detected by the EOG recording are mainly present during the Rapid Eye Movement (REM) sleep stage, while EMG shows variations in both its tonic and phasic levels depending on the sleep stage and the simultaneous presence of body movements.
[0006] A polysomnography provides various disadvantages. For instance, during an EEG electric potentials are recorded by using electrodes fixed on several sides of the skull, e.g. the electrodes are glued to the face and on the skull. Also, EOGs and EMGs require the attaching of electrodes and sensors to the face, skull or other body parts of the tested person. To detect eye movement during sleep, an EOG requires electrodes glued or otherwise attached near the eye or on the eyelid of the person.
[0007] All these electrodes further require wires that are attached to the electrodes and lead to a device placed near the head of the bed limiting the movement freedom of the tested person. Such recording is therefore obtrusive due to wiring, unusual sleep environments and imposed schedule at bedding conditions. The results of such tests may therefore be distorted due to the changed environment of the tested person. Other devices are bulky and uncomfortable or require a data recorder.
[0008] In addition, polysomnographies provide limitations due to the complexity of the recording techniques. In detail, specific recording places such as sleep laboratories and special equipment as well as well trained staff are necessary. Therefore, polysomnography remains an exceptional and expensive method for sleep evaluation.
[0009] A polysomnography system based on detection of eyelid movement (EOG), head movement and a heart beat signal (electrocardiogram - ECG) is described in US 5,902,250. The described system, however, is expensive and creates sleep disturbances
due to the number of sensors and wires necessary. In addition, the system described in US 5,902,250 is not very precise in determining sleep stages and does not determine sleep stage transitions.
[0010] WO 2012/156427 Al discloses a system and method for determining sleep, sleep stage and/or sleep stage transition of a person, including heart rate detecting means configured for detecting a heart rate of the person, movement detecting means configured for detecting a movement of a part of the body of the person, where the detected movement is caused by a skeletal muscle of the body, recording means configured for recording the detected heart rate and the detected movement of the part of the body, heart rate classifying means configured for classifying the recorded heart rate of the person into at least one heart rate class at least one heart rate variability class, movement classifying means configured for classifying the recorded movement into at least one movement class, and determining means configured for determining sleep, a sleep stage, a sleep stage transition and/or a sleep event of the person based at least partially on the at least one heart rate class and the at least one movement class.
Problem and solution
[0011] Two sensors of the same type may scatter and provide different signals even when measuring the same object or event. Further, a sensor used and sold with a respective signal receiving and processing device is different from the particular sensor used for developing a signal processing algorithm, here a sleep stage determining algorithm, run on the respective signal receiving and processing device. Further, the positioning of the sensor may affect its signals and thus the performance and accuracy of the sleep stage determining system.
[0012] The present invention aims to provide an improved sleep stage determining system.
[0013] The problem is solved by a sleep stage determining system according to claim 1
(first aspect) and by the methods according to claim 11 (second aspect) and claim 14 (third aspect). Preferred embodiments form the respective subject matter of the dependent claims.
[0014] The sleep stage determining system as described by claim 1 is suitable for determining a sleep stage and/or a sleep stage transition of a first person. The system comprises a sensor interface arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person’s cardiac activity. Further, the system comprises a sleep stage model unit arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
[0015] There is a risk that the first person does not position an acceleration sensor properly. Further, the acceleration sensor to be used by the first person may provide signals different from those of the acceleration sensor used to obtain the sleep stage model, particularly due to scatter. By relying of the cardiac activity signals and applying a sleep stage model to the received cardiac activity signals, the system dispenses with an acceleration sensor. Thereby the problem is solved.
[0016] The sleep stage model can be obtained by sensing cardiac activity signals of persons (probands) different from the first person, labelling different sleep stages in the obtained cardiac activity samples and subsequent machine learning using the labelled cardiac activity samples as training items. The sleep stage model can employ a similarity function k when processing the received cardiac activity signals. The sleep stage model can make use of pattern recognition or a support vector machine. The sleep stage model can be obtained by artificial intelligence including machine learning and/or rules from a sleep expert (expert rule).
[0017] The cardiac activity signals can be classified by a sleep stage model which has been trained on multiple training items, preferably by machine learning. A training item includes the cardiac activity signals which have been classified into the respective categories or labelled, for example, by an expert. The labels associated with the cardiac activity signals can be considered to represent ground truth.
[0018] Within the concept of the present invention a proband is a person who does not use the claimed system, in contrast to the first person (user). The proband’s cardiac activity is measured during a sampling interval and the obtained cardiac activity samples serve for training a sleep stage model to obtain a trained sleep stage model.
Preferred embodiments
[0019] The following preferred embodiments may be combined with each other, advantageously, unless indicated otherwise.
[0020] According to an embodiment the cardiac activity signal is an electrical heart signal, a pulse signal and/or a blood flow rate signal. The blood flow rate signal can be a blood volume flow rate signal. The cardiac activity signal maybe recorded using a singlelead electrocardiogram (ECG) or by monitoring the pulse rate through photoplethysmography (PPG) or ballistocardiography (BCG).
[0021] Another embodiment further comprises a wearable device which includes the sensor interface, preferably also the sleep stage model unit. This embodiment may be more convenient during use and/or can be used outside a sleep laboratory.
[0022] Alternatively, the sensor interface can be arranged in the wearable device together with a signal transmitting unit arranged for transmitting the cardiac activity signals to the sleep stage model unit located at a different location. The signal transmitting unit electrically connected to the sensor interface can be arranged to transmit the cardiac activity signals through a wire or wirelessly to the sleep stage model unit. This embodiment may be more convenient during use and/or can be used outside a sleep laboratory.
[0023] A further embodiment comprises the cardiac activity sensor which is connected to the sensor interface. The cardiac activity sensor can be selected from a light sensor for photoplethysmography, a pressure sensor for ballistocardiography and an electrode for a single-lead electrocardiogram. The cardiac activity sensor can be integrated in the wearable device.
[0024] In another embodiment, the sleep stage model unit is arranged to receive as input only the cardiac activity signals and/or arranged to apply the sleep stage model only to the received cardiac activity signals, to determine a sleep stage and/or a sleep stage transition. The embodiment can dispense with an acceleration sensor.
[0025] The following scientist demonstrated that sleep stages can be determined from just cardiac activity signals: a) Chen, Y.; Zhu, X.; Chen, W. Automatic sleep staging based on ECG signals using hidden Markov models. In Proceedings of the 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society
(EMBC), Milan, Italy, 25-29 August 2015; pp. 530-533; b) Lesmana, T.P.; Isa, S.M.; Surantha, N. Sleep Stage Identification Using the Combination of ELM and PSO Based on ECG Signal and HRV. In Proceedings of the 2018 3rd International Conference on Computer and Communication Systems (ICCCS), Nagoya, Japan, 27-30 April 2018; pp. 436-439-
[0026] The sleep stage model unit of a preferred embodiment is further arranged to apply a heart rate model to the received cardiac activity signals to determine heart rate data (S11), a heart rate variability and/or a heart rate event. The heart rate model can be arranged to determine a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or two subsequent R-peaks of the QRS complex (S11’). The heart rate model can be arranged to determine whether and by how much the time interval varies over time. The heart rate model can be arranged to calculate a heart rate average, a variability value (including a temporal and spectral indices), a rhythm characteristic, a heart rate acceleration and/or a heart rate event or change from the received cardiac activity signals. The heart rate model can be configured for classifying a sleep stage and/or a sleep stage transition based on the calculated heart rate average, instantaneous heart rate values, variability value, rhythm characteristic and/or heart rate event or change.
[0027] An embodiment comprises a sleep stage model unit which is arranged to receive as input the cardiac activity signals and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition, wherein the trained sleep stage model has been obtained by training a sleep stage model on multiple training items preferably by machine learning and/or artificial intelligence. The multiple training items comprise multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions. Supervised learning algorithms can be particularly suitable, such as classification, kernel machine, pattern recognition, regression, similarity learning, a support vector machine and transfer learning, among others. When two persons experience the same sleep stage, their cardiac activity signals may differ slightly. By employing multiple training items for training the sleep stage model, the trained sleep stage model may determine sleep stages and/or sleep stage transitions more accurately or reliably.
[0028] Preferably, the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing the cardiac activity of several probands, and by labelling different sleep stages in the
obtained cardiac activity samples. Each of the multiple training items includes the obtained cardiac activity samples and the labelled sleep stages. By employing multiple training items from several probands, the trained sleep stage model may determine sleep stages and/or sleep stage transitions more accurately or reliably.
[0029] The sleep stage model unit of another embodiment is arranged to receive as input the cardiac activity signals, and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition. The trained sleep stage model has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples and multiple body motility samples of probands, and labelled sleep stages and/or sleep stage transitions.
[0030] Preferably, the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing the cardiac activity of several probands, by measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the probands, and by labelling different sleep stages in the obtained cardiac activity samples, such that the multiple training items include the obtained cardiac activity samples, the obtained body motility samples, and the labelled different sleep stages.
[0031] An embodiment further includes a sampling device suitable for obtaining multiple training items during the sampling interval, the sampling device comprising a cardiac activity sensor for sensing a proband’s cardiac activity signals. Preferably, the cardiac activity sensor of the sampling device is of the same type as the cardiac activity sensor of the above embodiments. The sampling device can further comprise a body motility sensor for sensing the proband’s body motility. The body motility sensor can be an acceleration sensor, a pressure sensor or a camera.
[0032] A method of training a sleep stage model to be used with any of the above systems forms a second aspect of the invention. The method of training a sleep stage model comprises the steps:
51 measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing a cardiac activity of several probands,
52 labelling different sleep stages and/or sleep stage transitions in the obtained cardiac activity samples, thereby obtaining multiple training items comprising
multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions,
53 training the sleep stage model on the multiple training items, preferably by machine learning, thereby obtaining the trained sleep stage model.
Step Si can further include measuring body motility. The labelling during step S2 can be based on polysomnography, either by “traditional manual” scoring by a sleep expert or a trained physician, or by automatic scoring using a validated algorithm. The labelled sleep stages and/or sleep stage transitions of the multiple training items can be considered to represent ground truth.
This method may help to improve the accuracy or reliability of determining sleep stages and/or sleep stage transitions.
[0033] A preferred method of training a sleep stage model further comprises:
54 measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the several probands, wherein the multiple training items comprise the multiple cardiac activity samples and multiple body motility samples of probands and associated sleep stages and/or sleep stage transitions.
[0034] A method of determining a sleep stage and/or a sleep stage transition of the first person (third aspect) employs one of the systems explained above. The method comprises the steps:
S10 obtaining cardiac activity signals from the cardiac activity sensor sensing the first person’s cardiac activity signals,
S21 applying the sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
Preferably, the sleep stage model of step S21 is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
[0035] A preferred method of determining a sleep stage and/or a sleep stage transition of the first person comprises above step Sio, obtaining heart rate data from the cardiac activity signals (Sil), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (Sil’), and applying the sleep stage model to the heart rate data to determine a sleep stage and/or a sleep stage transition (S21’). Preferably, the sleep stage model of step S21’ is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples or heart rate data of probands and associated sleep stages and/or sleep stage transitions.
[0036] Another method of determining a sleep stage and/or a sleep stage transition of the first person employs one of the systems explained above. Cardiac activity signals are obtained from the cardiac activity sensor sensing the first person’s cardiac activity signals (Sio). Heart rate data is obtained from the cardiac activity signals (S11), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (S11’). The heart rate data can be prefiltered and abnormalities detected (S12). The prefiltered heart rate data can be parameterised by filters, FFT or the like (S13). The parameterised signals can be used with a support vector machine (S14) and sleep stage probabilities can be obtained. The prefiltered heart rate data or heart rate signals can be examined for detecting physiological events (S15), such as cardiac arousals (AC), small-AC-like-Patterns (Salp) and variations in heart rate. Step S15 may be done through the use of pattern recognition and signal processing techniques, among others. The detected physiological events can be used for estimating body motility (S16), as movements have direct effects on the cardiac activity. The sleep stage probabilities, the detected physiological events and estimated body motility can contribute to the sleep stage model (S17). The sleep stage model includes sleep stage transition rules and permits to determine sleep stage transitions by these transition rules. The sleep stage model can include a) interpreting an increased estimated body motility and increased cardiac activity events as a sleep stage transition towards wakefulness, b) interpreting a reduced estimated body motility and reduced cardiac activity events as a sleep stage transition towards deep Non-REM sleep. The sleep stage model can determine a subsequent sleep stage from the present sleep stage if the right conditions are fulfilled among others, based on probability.
The sleep stages can be classified with the sleep stage transitions (S18) and can be shown in a hypnogram. Steps S12 to S18 can be considered as details of step S21.
[0037] A sleep stage classification can be performed by applying the above sleep stage transition rules. Beginning at a first sleep state, which can be a wake state at the beginning of the night, the sleep stage model applies the sleep stage transition rules on the received cardiac activity signals for obtaining the most probable sleep stage transition (S25). The sleep stage transition rules permit the sleep stage model unit to compute the time of each possible transition applicable from the current sleep state, based on detected physiological events, pre-detected sleep events and previously determined sleep stages (S25). From the first sleep stage the most probable sleep stage transition leads to a subsequent sleep stage (S26). The next possible and most probable sleep stage transition at point in time Tn indicates the next sleep state (S26). Steps S25, S26 can be repeated along the received cardiac activity signals until the end of the night (“lights on”) is reached or no more sleep stage transitions are found. Before repeating steps S25, S26, heart rate data can be obtained from the cardiac activity signals (S11), such that S25, S26 are performed on heart rate data.
Exemplary embodiments
[0038] Further features and advantages are apparent to the skilled person from the following exemplary embodiments.
[0039] During sleep, several sleep stages can be determined. These sleep stages can be categorized as rapid eye movement (REM) sleep stage and non-REM sleep stages. The REM sleep stage is the one where vivid dreaming occurs. It can be identified by the occurrence of rapid eye movements under closed eyelids, motor atonia and low voltage EEG patterns. The REM sleep stage, also referred to as REM sleep, is also associated with bursts of muscular twitching, irregular breathing, irregular heart rate and increased autonomic activity. Periods of REM sleep are also referred to as paradoxical sleep. Moreover, the sleep of a person can also be scored into non-REM (NREM) stages, which are numbered 1 to 3.
[0040] Figure 1 depicts an exemplary hypnogram of a healthy young adult showing the different sleep stages of an eight hour sleep recording. It must be noted that the transitions from one stage to another are conventionally considered as abrupt steps. As illustrated,
within the first hour of sleep a person starting from a wakeful state and falling asleep may transit to NREM sleep stage 1 and further to stages 2, and 3.
[0041] The criteria of sleep stage 1 of the NREM sleep consist of a low voltage EEG tracing with well defined alpha activity and theta frequencies in the 3 to 7 Hz range, occasional vertex spikes, and slow rolling eye movements (SEMs). This stage includes the absence of sleep spindles, K- complexes and REMs. Stage 1 normally represents 4 to 5% of the total amount of sleep time.
[0042] The sleep stage 2 of NREM sleep is characterized by the occurrence of sleep spindles and K- complexes against a relatively low voltage, mixed frequency EEG background. High voltage delta waves may comprise up to 20% of stage 2 epochs. The sleep stage 2 usually accounts for 45 to 55% of the total sleep time.
[0043] A light Non-REM sleep stage is a common term for the sleep stages 1 and 2 (light sleep), while a deep Non-REM sleep is a term for sleep stage 3.
[0044] Returning to Figure 1, after a period of sleep stage 3 the sleep of the tested person changes to sleep stage 2 and to REM sleep. Further, a phase of light Non-REM sleep stages follow to then return to another deep Non-REM sleep stage.
[0045] The rest of the sleep as depicted in Figure 1 comprises transitions from REM sleep periods to lighter Non-REM sleep stages, such as stages 1 and 2.
[0046] In order to determine the sleep stages and/or the sleep stage transitions of the tested first person, the present invention provides a system for sensing and recording continuously and up to several days or weeks the basic physiological variables such as heart rate and body motility together with some characteristics of the ambient physical environment. This methodology will be able to score the basic states such as waking and sleeping periods of the tested person. During the sleep state, sleep stages will be scored every 30-second epoch. Moreover, the simultaneous recording of ambient physical variables together with the biological ones will allow evaluating the possible impact of the former to the latter.
[0047] Figure 2 shows an exemplary sleep stage determining system too for determining a sleep stage and/or a sleep stage transition of a first person. The system comprises a sensor interface 101 arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person’s cardiac activity, and a sleep stage model unit 102
arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
[0048] The sensor interface and the sleep stage model unit can be arranged in a wearable device 106 (dashed line). The exemplary system can comprise a cardiac activity sensor 105 (dashed line) electrically connected to the sensor interface 101. The cardiac activity sensor can form a part of the wearable device. Alternatively, the sleep stage model unit can be at a different location.
[0049] The sensor interface of another exemplary sleep stage determining system (not shown) is connected with a cardiac activity sensor but with no other sensor.
[0050] A sleep stage model can be obtained by an exemplary method comprising sensing cardiac activity signals of probands during a sampling interval (Si) and labelling different sleep stages in the measured cardiac activity signals or cardiac activity samples (S2). The sleep stage model can employ a similarity function k when processing the received cardiac activity signals particularly of probands. Obtaining the sleep stage model can involve pattern recognition applied to the cardiac activity signals of several probands.
[0051] A sleep stage model can be obtained by another exemplary method comprising steps Si, S2 and by training a sleep stage model on multiple training items , preferably by machine learning (S3), see figure 3, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
[0052] A sleep stage model can be obtained by a further exemplary method comprising steps Si, S2 and by training a sleep stage model on multiple training items, preferably by machine learning (S3), the multiple training items comprising multiple cardiac activity samples and multiple body motility samples of probands, and labelled sleep stages and/or sleep stage transitions. The body motility samples of probands are obtained by measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the probands (S4), see figure 3.
[0053] An exemplary method of determining a sleep stage and/or a sleep stage transition of the first person employs one of the systems explained above. The exemplary method is shown in figure 4a and comprises the steps:
S10 obtaining cardiac activity signals from the cardiac activity sensor sensing the first person’s cardiac activity signals,
S21 applying the sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
Preferably, the sleep stage model of step S21 is a trained sleep stage model, which has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
[0054] An exemplary method of determining a sleep stage and/or a sleep stage transition of the first person employs one of the systems explained above. Cardiac activity signals are obtained from the cardiac activity sensor sensing the first person’s cardiac activity signals (S10). Heart rate data is obtained from the cardiac activity signals (S11), particularly by repeatedly determining a time interval between two subsequent peak blood pressures, two subsequent peak blood flow rates or between two subsequent R-peaks of the QRS complex for the heart rate data (S11’). The heart rate data can be prefiltered and abnormalities detected (S12). The prefiltered heart rate data can be parameterised by filters, FFT or the like (S13). The parameterised signals can be used with a support vector machine (S14) and sleep stage probabilities can be obtained. The prefiltered heart rate data or heart rate signals can be examined for detecting physiological events (S15), such as cardiac arousals (AC), small-AC-like-Patterns (Salp) and variations in heart rate. Step S15 maybe done through the use of pattern recognition and signal processing techniques, among others. The detected physiological events can be used for estimating body motility (S16), as movements have direct effects on the cardiac activity. The sleep stage probabilities, the detected physiological events and estimated body motility can contribute to the sleep stage model (S17). The sleep stage model includes sleep stage transition rules and permits to determine sleep stage transitions by these transition rules. The sleep stage model can include a) interpreting an increased estimated body motility and increased cardiac activity events as a sleep stage transition towards wakefulness, b) interpreting a reduced estimated body motility and reduced cardiac activity events as a sleep stage transition towards deep Non-REM sleep. The sleep stage model can determine a subsequent sleep stage from the present sleep stage if the right conditions are fulfilled among others, based on probability. The sleep stages can be classified with the sleep stage transitions (S18), and can be shown in a hypnogram. The exemplary method is shown in figure 4b.
[0055] A sleep stage classification can be performed by applying the above sleep stage transition rules. Beginning at a first sleep state, which can be a wake state at the beginning of the night, the sleep stage model applies the sleep stage transition rules on the received cardiac activity signals for obtaining the most probable sleep stage transition (S25). The sleep stage transition rules permit the sleep stage model unit to compute the time of each possible transition applicable from the current sleep state, based on detected physiological events, pre-detected sleep events and previously determined sleep stages (S25). From the first sleep stage the most probable sleep stage transition leads to a subsequent sleep stage (S26), see figure 6. The next possible and most probable sleep stage transition at point in time Tn indicates the next sleep state (S26), see figure 5. Steps S25, S26 can be repeated along the received cardiac activity signals until the end of the night (“lights on”) is reached or no more sleep stage transitions are found, see figures 5, 6. Before repeating steps S25, S26, heart rate data can be obtained from the cardiac activity signals (S11), such that S25, S26 are performed on heart rate data, see figure 6.
Claims
1. A sleep stage determining system for determining a sleep stage and/or a sleep stage transition of a first person, wherein the system comprises: a sensor interface arranged to obtain cardiac activity signals from a cardiac activity sensor sensing the first person’s cardiac activity, and a sleep stage model unit arranged to receive as input the cardiac activity signals and to apply a sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
2. The system according to claim 1, wherein the cardiac activity signal is an electrical heart signal, a pulse signal and/or a blood flow rate signal.
3. The system according to any preceding claim, further comprising the cardiac activity sensor which is connected to the sensor interface.
4. The system according to any preceding claim, further comprising a wearable device which includes the sensor interface and the sleep stage model unit.
5. The system according to any preceding claim, wherein the sleep stage model unit is arranged to receive as input only the cardiac activity signals and/or arranged to apply the sleep stage model only to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition.
6. The system according to any preceding claim, wherein the sleep stage model unit is further arranged to apply a heart rate model to the received cardiac activity signals to determine a heart rate, a heart rate variability and/or a heart rate event.
7. The system according to any preceding claim, wherein the sleep stage model unit is arranged to receive as input the cardiac activity signals and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition, wherein the trained sleep stage model has been obtained by training a sleep stage model on multiple training items, preferably by machine learning and/or artificial intelligence, the multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions.
8. The system according to claim 7, wherein the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a heart cardiac activity sensor sensing the cardiac activity of several probands, and by labelling different sleep stages and/or sleep stage transitions in the obtained cardiac activity samples, such that the multiple training items include the obtained cardiac activity samples and the labelled sleep stages.
9. The system according to one of claims 1 to 6, wherein the sleep stage model unit is arranged to receive as input the cardiac activity signals and to apply a trained sleep stage model to the received cardiac activity signals to determine a sleep stage and/or a sleep stage transition, wherein the trained sleep stage model has been obtained by training a sleep stage model on multiple training items, preferably by machine learning, the multiple training items comprising multiple cardiac activity samples and multiple body motility samples of probands, and labelled sleep stages and/or sleep stage transitions.
10. The system according to claim 9, wherein the multiple training items have been obtained by measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing the cardiac activity of several probands, by measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the several probands, and by labelling different sleep stages in the obtained cardiac activity samples, such that the multiple training items include the obtained cardiac activity samples, the obtained body motility samples, and the labelled sleep stages.
11. The system according to any preceding claim, further including a sampling device suitable for obtaining multiple training items during a sampling interval, the sampling device comprising a cardiac activity sensor for sensing a proband’s cardiac activity signals, preferably further comprising a body motility sensor for sensing the proband’s body motility.
12. Method of training a sleep stage model to be used with a system according to any of the preceding claims, comprising the steps:
Si measuring cardiac activity signals during a sampling interval from a cardiac activity sensor sensing a cardiac activity of several probands,
52 labelling different sleep stages and/or sleep stage transitions in the obtained cardiac activity samples, thereby obtaining multiple training items comprising multiple cardiac activity samples of probands and associated sleep stages and/or sleep stage transitions,
53 training a sleep stage model on the multiple training items, preferably by machine learning, thereby obtaining the trained sleep stage model. Method according to claim 12, further comprising
54 measuring body motility signals during the sampling interval from a body motility sensor sensing the body motility of the several probands, wherein the multiple training items comprise the multiple cardiac activity samples and the multiple body motility samples of probands and associated sleep stages and/or sleep stage transitions. Method of determining sleep stages and/or a sleep stage transitions of a first person, using a system according to any of claims 1 to 11, the method comprising:
510 obtaining cardiac activity signals from the cardiac activity sensor sensing the first person’s heart rate activity,
511 obtaining heart rate data from the cardiac activity signals.
17
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP21193315.5 | 2021-08-26 | ||
EP21193315 | 2021-08-26 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023025770A1 true WO2023025770A1 (en) | 2023-03-02 |
Family
ID=77838663
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/EP2022/073417 WO2023025770A1 (en) | 2021-08-26 | 2022-08-23 | Sleep stage determining system |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2023025770A1 (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150190086A1 (en) * | 2014-01-03 | 2015-07-09 | Vital Connect, Inc. | Automated sleep staging using wearable sensors |
US20170055898A1 (en) * | 2015-08-28 | 2017-03-02 | Awarables, Inc. | Determining Sleep Stages and Sleep Events Using Sensor Data |
WO2018048951A2 (en) * | 2016-09-06 | 2018-03-15 | Fitbit, Inc. | Methods and systems for labeling sleep states |
EP3454743A1 (en) * | 2016-06-27 | 2019-03-20 | Koninklijke Philips N.V. | Determination system and method for determining a sleep stage of a subject |
-
2022
- 2022-08-23 WO PCT/EP2022/073417 patent/WO2023025770A1/en unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150190086A1 (en) * | 2014-01-03 | 2015-07-09 | Vital Connect, Inc. | Automated sleep staging using wearable sensors |
US20170055898A1 (en) * | 2015-08-28 | 2017-03-02 | Awarables, Inc. | Determining Sleep Stages and Sleep Events Using Sensor Data |
EP3454743A1 (en) * | 2016-06-27 | 2019-03-20 | Koninklijke Philips N.V. | Determination system and method for determining a sleep stage of a subject |
WO2018048951A2 (en) * | 2016-09-06 | 2018-03-15 | Fitbit, Inc. | Methods and systems for labeling sleep states |
Non-Patent Citations (2)
Title |
---|
CHEN, Y.ZHU, X.CHEN, W.: "Automatic sleep staging based on ECG signals using hidden Markov models", PROCEEDINGS OF THE 2015 37TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC), MILAN, ITALY, 25 August 2015 (2015-08-25), pages 530 - 533, XP032810198, DOI: 10.1109/EMBC.2015.7318416 |
LESMANA, T.P.ISA, S.M.SURANTHA, N: "Sleep Stage Identification Using the Combination of ELM and PSO Based on ECG Signal and HRV", PROCEEDINGS OF THE 2018 3RD INTERNATIONAL CONFERENCE ON COMPUTER AND COMMUNICATION SYSTEMS (ICCCS), NAGOYA, JAPAN, pages 436 - 439 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9820680B2 (en) | System and method for determining sleep and sleep stages of a person | |
JP6721156B2 (en) | Biological information analysis device, system, and program | |
Paalasmaa et al. | Unobtrusive online monitoring of sleep at home | |
JP6516846B2 (en) | Device and method for sleep monitoring | |
CN105142515B (en) | The method for determining the mankind's sleep stage for being conducive to waking up | |
US9833184B2 (en) | Identification of emotional states using physiological responses | |
EP2185063B1 (en) | Method and system for monitoring sleepiness | |
CN109328034B (en) | Determining system and method for determining sleep stage of subject | |
US20090024044A1 (en) | Data recording for patient status analysis | |
JP2018524080A (en) | Apparatus and method for monitoring the physiological state of a subject | |
EP3536225A1 (en) | Sleep apnea detection system and method | |
WO2003038649A1 (en) | Improvements in or relating to physiological monitoring | |
JP7301275B2 (en) | Sleep state estimation system | |
WO2023025770A1 (en) | Sleep stage determining system | |
Gerla et al. | Multichannel analysis of the newborn EEG data | |
Lee et al. | Monitoring obstructive sleep apnea with electrocardiography and 3-axis acceleration sensor | |
EP4088651A1 (en) | Method for detecting sleep disordering events | |
US20230157632A1 (en) | Detecting Obstructive Sleep Apnea/Hypopnea Using Micromovements | |
KR102569058B1 (en) | Patch type polysomnography apparatus and method | |
JP2004514491A (en) | Method and apparatus for diagnosing sleep apnea in a patient awake state | |
JP6775359B2 (en) | How to operate the nausea detector and the nausea detector | |
Gupta et al. | Overview of the effect of physiological stress on different biological signals | |
Baek et al. | Computer-aided detection with a portable electrocardiographic recorder and acceleration sensors for monitoring obstructive sleep apnea | |
Bertschi | Optical wrist-worn device for sleep monitoring |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22768418 Country of ref document: EP Kind code of ref document: A1 |