Patent Application
20250213124
This application claims benefit of UK Patent Application No. 2400080.4 entitled, “System and Methods for Improving Sleep Quality” filed Jan. 3, 2024, the entire disclosure of which is incorporated herein by reference.
The present disclosure generally relates to a method of controlling a sleep enhancing device or system, for improving sleep quality for a sleeping person; a system for carrying out that method; and a method of generating, by a machine learning model, control data for controlling a sleep enhancing device or system for improving sleep quality.
Sleep is essential for everyone, but it does not always come easily. A person may have difficulty getting to sleep, difficulty staying asleep, difficulty waking up, difficulty getting back to sleep after waking up, or may still feel tired after waking up, or a combination of any of these.
Sleep interruptions may occur sporadically due to an illness or as a result of snoring, for example, but disrupted sleep resulting from a sleep disorder can be extremely debilitating and can affect all areas of a person's life. There are over 80 different sleep disorders known at present.
Other people may have no trouble sleeping per se, but may have busy personal and/or working lives that place many demands on their time. Most people require 7-8 hours of sleep per day but may not regularly achieve that amount of sleep, leaving them with sleep debt that can accumulate and may be debilitating in its own way.
Somnology is the scientific study of sleep and may be used to help people who experience disrupted sleep, particularly in relation to sleep disorders. Somnology divides sleep into four distinct stages/periods, referred to as sleep stage 1 (N1), sleep stage 2 (N2), sleep stage 3 (N3) and sleep stage 4 (REM), which are discussed further below.
During sleep, there is a progressive passage through the stages of sleep from stage 1 (N1) to sleep stage 4 (REM), which forms a sleep cycle, after which the cycle resumes from sleep stages 2 to 4. A complete sleep cycle in adults lasts between 90 to 110 minutes. The first or initial sleep cycles include relatively short periods of REM sleep and relatively long periods of deep sleep. During the night, the periods of REM sleep increase in length, and those of deep sleep become shorter. A person who is asleep for a night will generally have four to six sleep cycles, meaning that they experience the various sleep stages several times.
Considering the sleep stages in more detail, the first sleep stage is stage 1, which is considered to be ‘light’ sleep where the body has not fully relaxed yet. Sleep stage 1 (N1) can last about 5 to 10 minutes and represents the beginning of the transition from wakefulness to sleep.
The next sleep stage 2 (N2) can last for 10 to 25 minutes during the first sleep cycle (i.e. the first instance of sleep stage 2 after having fallen asleep). Each sleep stage 2 can become longer during the course of the person's time asleep. In total, a person typically spends about 50% of their total sleep in sleep stage 2. Eye movements stop and brain waves slow down more during sleep stage 2.
Sleep stage 3 (N3) is deep sleep and represents approximately 20% of the total sleep time. Sleep stage 3 is the most important for rest and regeneration. In this phase, no muscle activity is registered, and the brain emits slow brain waves, called delta waves, associated with states of rest and deep regeneration. A healthy person spends the longest time in sleep stage 3 during the first half of the night. During the initial sleep cycles, sleep stage 3 usually lasts for 20 to 40 minutes. Throughout the night, instances of sleep stage 3 become shorter and more time is spent in stage 4 instead.
Sleep stage 4 is REM sleep (where REM means rapid eye movement) and represents approximately 25% of the total sleep time. Sleep stage 4 is the stage in which dreams occur. Sleep stage 4 plays a role in strengthening memory, processing emotions, and general regeneration of the body.
During sleep stage 4 (REM), the brain functions in many ways that are similar to when awake. The brain waves are active, and the breathing and heart rate increase and become more variable than in other sleep stages, which implies that the body does not rest during the REM sleep. If the amount of time spent in REM sleep during the night is too long, which implies a lack of deep sleep in sleep stage 3, the subject will feel tired in the morning.
Although REM sleep was previously believed to be the most important sleep phase for learning and memory, newer studies have suggested that non-REM sleep (that is, stages 1, 2 and 3) are more important in this respect, leaving aside the advantage that these stages are also more restful and restorative than REM sleep. Given that there is a surge in activity during REM sleep, this may explain why already vulnerable people often experience heart attacks and other events in the early morning hours, which is typically spent more in REM sleep.
There are scientific studies that have shown applying a slow oscillating movement to a bed reduces the time taken to initiate sleep, as well as increasing the time spent in a deeper state of sleep, and can lead to improvements in a person's memory capabilities. The Applicant's patent application WO2018083153 discloses a method for controlling a rocking bed where the bed rocking motion is stopped if the user leaves the bed or if the user falls asleep (e.g. because they have stopped moving).
However, each person has a different sensitivity to rocking or oscillatory motion. If this movement is in excess for a given person, there is a risk that that person will feel dizzy or even sick or nauseous. This is far from ideal for a relatively fit and healthy person, and may present a real danger to an elderly or more vulnerable person.
It is an object of the present disclosure to reduce or substantially obviate the aforementioned problems.
According to a first aspect of the present disclosure, there is provided a method of controlling a sleep enhancing device or system (e.g. on or next to a bed), for improving sleep quality for a person, the method comprising the steps of:
The disclosure has major advantages for improving overall sleep quality, such as allowing a person to have more restorative sleep. The disclosure can prolong sleep duration in a particular sleep state and/or reduce sleep duration in another particular sleep state. That is, the time spent in one sleep stage can be controlled relative to the time spent in another sleep stage.
Put another way, the disclosure allows for the time spent in deep sleep and/or REM to be maximized. In particular, the disclosure can allow a person to have a controllably reduced proportion of their time spent asleep whilst in sleep stage 1 (N1) and sleep stage 2 (N2) relative to sleep stage 3 (N3) and/or sleep stage 4 (REM). For example, the disclosure may allow a person to achieve or reach sleep stage 3 (N3) more quickly than would otherwise occur.
The method aims to maintain a state of deep sleep for the bed user, while minimizing possible adverse effects experienced due to the bed oscillatory motion. That is, the bed motion is tailored to the person and/or controlled (preferably automatically) in a way that depends on the sleep stage that the person is estimated to be in.
The instruction or decision to start and/or stop and/or adjust the input provided by the device/system is preferably determined by an artificial intelligence (AI) algorithm or machine learning model that has been trained on a suitable number of data sets. In general, the larger the number of data sets used for training, the better the training outcome for the resulting model. The number of data sets may be about at least about 10000, for example.
Based on information provided by sensors which are sensing or periodically/continuously monitoring vital parameters of the person's body, the algorithm or model can precisely estimate of the sleep stage that the person is in and provide an instruction/decision accordingly.
The disclosure may utilize AI algorithm(s) which, based on information received from sensors during a person's time asleep, can learn the patterns and/or sleep stages of a person's sleep cycles (for example based on data acquired over several nights of that person's sleep). Based on that information, the AI algorithm(s) may then make predictions and determinations as to when an input from a sleep enhancing device/system ought to be started and/or stopped and/or adjusted for improving that person's sleep quality.
The disclosure is suitable for any person who suffers from a sleep disorder and/or from any secondary health effects that have arisen or worsened due to a lack of sleep.
The disclosure is also suitable for any person who does not have any sleep problems/disorders, for example if they wish to reduce their total required sleep time. That is, to achieve a restful sleep in a shorter time.
In preferred embodiments, the time spent by a sleeping person in sleep stages 1 and 2 should be reduced to a minimum, whilst the time spent in sleep stage 3 should be quickly entered, maintained for longer than normal and alternated with sleep stage 4 (REM) which is important for the memory regeneration.
The method may therefore prioritize initiation and/or maintenance of sleep stage 3 and provide input(s) to achieve this. The method may also or instead prioritize artificially lengthened and/or artificially ended periods for either or both of sleep stages 3 and 4 for controlled transfer of the person between sleep stages 3 and 4, whilst minimizing time spent in sleep stages 1 and 2, by providing input(s) to achieve this.
Estimates of which particular sleep stage exists at the time of a given sensor reading may be based on confidence scores/values. That is, the sensor reading may be processed or assessed to determine the likelihood that it corresponds to sleep stage 3 (N3) or sleep stage 4 (REM) or another sleep stage or stages, and the sleep stage which has the corresponding highest confidence score/value may be determined as the estimated sleep stage.
Estimates may be made periodically or substantially continuously. For example, the frequency at which estimates are made may depend on the frequency of data sampling by the sensors and/or a user-set value for how frequently the sleep stage estimation should be refreshed or re-estimated.
The sleep enhancing device or system may be a bed rocking device or system. For example, the bed rocking may be achieved using a device/system similar to that presented in the Applicant's patent application WO2018083153 or the Applicant's patent application EP4137008.
The input may be a bed rocking motion or a reciprocating motion or an oscillatory motion. Any input value discussed below may correspond to a bed rocking motion speed or a reciprocating motion speed or an oscillatory motion speed. The input value in such cases could vary between 0 (i.e. stopped or at rest) to a maximum value of 100 (which may correspond to a speed of approximately 10 mm per second).
If the person is estimated to be in either or both of sleep stage 1 (N1) and sleep stage 2 (N2), the method may include operating (or continuing to operate) the sleep enhancing device(s) or system(s). That is, rather than starting/stopping/adjusting the input, the method may allow the sleep enhancing device/system to continue at its previous input value or in its current state.
The input during sleep stages 1 and/or 2 may be a non-zero input. For example, a bed rocking motion or bed speed may have a non-zero value during any estimated sleep stage 1 and/or sleep stage 2 during the method.
In other words, the bed may move or rock during sleep stage 1 and/or sleep stage 2. The input value in sleep stage 1 and/or sleep stage 2 may be a predetermined or user-set value. That is, for operation of the sleep enhancing device in sleep stages 1 and/or 2 the input may have a value preset according to user choice, and for operation of the sleep enhancing device in sleep stages 3 and/or 4 the input value may be set or varied according to control data which may be or have been generated by a machine learning model.
Rocking the bed up to a certain speed during sleep stages 1 and/or 2 can help the person to reach sleep stage 3 (N3) more quickly, at which point the bed rocking motion may be reduced or ceased.
The method may include estimating, based on information from at least the one or more sensors, when the person enters or commences sleep stage 3 (N3) and/or sleep stage 4 (REM). The method may include stopping or adjusting the input provided by the sleep enhancing device or system when the person is estimated to have entered sleep stage 3 (N3) or sleep stage 4 (REM).
In other words, when the user is in a deep sleep, the movement of the bed is preferably reduced (optionally gradually reduced) to a minimum or stopped. This can reduce possible adverse effects resulting from the sleep enhancing device or system, such as adverse effects from an oscillatory/rocking movement.
The method may include estimating whether sleep stage 3 (N3) and/or sleep stage 4 (REM) has elapsed by a predetermined extent. The method may include starting or adjusting (optionally gradually starting/adjusting) the input provided by the sleep enhancing device or system when the relevant sleep stage is estimated to have elapsed by the relevant predetermined extent.
This can help to prolong sleep stage 3 and/or end sleep stage 4, relative to what would otherwise happen for the person during non-augmented sleep.
The predetermined extent for sleep stage 3 (N3) may correspond to a few minutes (perhaps up to 10 minutes) before a predicted end of that sleep stage 3.
The predicted end may be based on a determined success rate of previous input starts prior to the end of the sleep stage 3. For example, an input (re)start during sleep stage 3 may occur incrementally or progressively earlier, that is prior to the predicted end of sleep stage 3, in subsequent sleep stages 3 if the input start of a previous sleep stage does not have the desired effect, e.g. prolonging that sleep stage 3.
The sleep enhancing device or system may be programmable with a predetermined input value for sleep stage 3 (N3). The input generated or conveyed by the sleep enhancing device/system may be set at the predetermined input value (optionally gradually incremented to that value) at around the predicted end of the sleep stage 3 (N3).
This allows for a user-defined input value such as bed rocking speed to be realized according to the user's sensitivity to the input, that is what they are comfortable with.
Any predetermined input value may be set via a remote control (which may be provided with the sleep enhancing device/system) and/or via an app on an electronic device (e.g. smartphone or tablet) programmed to communicate with the sleep enhancing device/system.
Where the person is estimated to be in sleep stage 3 (N3), the input may stopped if the duration of that sleep stage 3 (N3) exceeds about 60 minutes or exceeds a predetermined length of time. The predetermined length of time may be a user set value or may be a clinically set value (e.g. determined by a doctor), which may be greater than or less than 60 minutes.
This avoids prolonging sleep stage 3 excessively, in order to achieve an optimum augmented balance with the other sleep stages, and may facilitate alternation with sleep stage 4.
If the input is stopped at the end of sleep stage 3 (N3) and the person next enters sleep stage 4 (REM), then the input may (for a time) be maintained at a zero input value.
If the predetermined input value was insufficient to prolong a first sleep stage 3 (N3) period, then a higher input value may be used during one or more subsequent sleep stage 3 (N3) periods. The input value may be about 5% higher than the predetermined value.
The sleep enhancing device or system may be programmable with a predetermined input start time for sleep stage 3 (N3). The input may start at the predetermined input start time when the person is estimated to be in sleep stage 3 (N3).
These features can variously help to prolong a sleep stage 3 for a desired longer duration in order to provide a beneficially longer period of deep sleep.
The predetermined extent for sleep stage 4 (REM) may correspond to about half of the predicted duration of that sleep stage 4 (REM).
The sleep enhancing device or system may be programmable with a predetermined input value for sleep stage 4 (REM). The input may be set at the predetermined input value at around midway through the sleep stage 4 (REM).
If the predetermined input value was insufficient to end a first sleep stage 4 (REM) period, then a higher input value may be used during one or more subsequent sleep stage 4 (REM) periods. The input value may be about 5% higher than the predetermined value.
The sleep enhancing device or system may be programmable with a predetermined input start time for sleep stage 4 (REM). The input may start at the predetermined input start time when the person is estimated to be in sleep stage 4 (REM).
These features can variously help to controllably end a sleep stage 4 earlier than during non-augmented sleep, and may help facilitate an earlier return to a subsequent sleep stage 3 for the person during a given sleep.
The one or more other sleep stages may include sleep stage 1 (N1) and/or sleep stage 2 (N2). When the person is estimated to be in sleep stage 1 (N1) and/or when the person is estimated to be in sleep stage 2 (N2), the method may include performing at least one of:
A person may enter various different sleep stages at various different times during the course of a sleep. It will be appreciated that the method may involve control of the input (whether starting, stopping, maintaining and/or adjusting the input) at various different points of some or all of those sleep stages, which may include control of the input at the commencement of a particular sleep stage (according to an estimate made during the method) and/or control of the input partway through a particular sleep stage (again according to an estimate made during the method) and/or control at the end of a particular sleep stage (again according to an estimate made during the method).
It will also be appreciated that control of the input during a sleep stage may include multiple instances of controlling the input during the course of a single sleep stage.
The sleep enhancing device or system may be programmable with clinical information for the person. Any one or more of the following input parameters may be controlled based at least in part on the clinical information: input type, input intensity, input start time, input end time, input duration, input pattern.
A doctor or other medical/care professional can provide the relevant information to the device/system, which may affect its operation. For example, if the person has a diagnosed sleep condition/disorder then this overrides or takes precedence over what the control data or instructions may contain.
Clinical information may be used by the device/system in a way that seeks to avoid an input exacerbating an existing condition and/or seeks to minimize the likelihood of a condition being triggered. For example, if a user has a cardiovascular-related problem, and might suffer a heart attack during REM sleep, the device/system may reduce the time for which bed rocking is halted during sleep stage 4.
The sleep enhancing device or system may be programmable with a sleep duration value. The sleep duration value may be measured beginning from a clock time or a time when the person is detected as having fallen asleep.
Where a given sleep stage 3 (N3) or sleep stage 4 (REM) is predicted to be the last sleep stage 3 (N3) or last sleep stage 4 (REM) before the sleep duration value is exceeded, the sleep enhancing device or system and the associated input may be stopped and may not restart when the person enters any subsequent sleep stage before they are awake.
The person can thus set a time for which they would like to be asleep and the sleep enhancing device/system operates accordingly. Stopping the input (such as bed rocking) during the last of the predicted deep sleep stages, and not restarting it in subsequent sleep stages prior to waking, can help to avoid making the person feel dizzy when they wake up.
Any suitable sensor may be used. A consumer or medical device (or devices) may integrate sensors for monitoring vital parameters which provide information about the sleep stages. Examples of such devices include any or more of the following for provision in the sleep enhancing system.
A bracelet or wristwatch may be provided which includes sensor(s) for any one, some or all of: motion, temperature, measure heart rate/pulse, electrocardiogram (ECG), blood pressure.
A chest belt may be provided which includes sensor(s) for any one, some or all of: movement, temperature, heart rate/pulse measurement, electrocardiogram (ECG), heart rate variability.
A head band may be provided which includes sensor(s) for any one, some or all of: motion, temperature, and EEG (electro encephalogram).
A mat-type sensor may be provided for placement on or under a mattress. The mat-type sensor may include sensor(s) for any one, some or all of: motion, breathing rate, heart rate/pulse, heart rate variability.
An accelerometer type sensor may be provided for positioning on the mattress for sensing motion. The purpose of this sensor is to sense movement of the person on the mattress, which may happen when the person on the mattress changes their lying position, for example by turning over. The accelerometer type sensor may not be used to sense a lateral movement of the mattress caused by a bed rocking apparatus or device, for example.
All of the above sensors have various advantages and disadvantages due to their positioning in relation to the person's body during sleep. Some of them may be uncomfortable during sleep (such as the chest belt or head band). Whichever device(s) are provided and worn by a person during sleep, the positioning of each device in relation to the person's body and the type of device used affects the accuracy and hence utility of the readings taken of the person's vital parameters.
Since there is presently no universal sensor that suitably senses all of the body's vital parameters, using a combination of multiple sensors (preferably 3-6 or more different sensors) that the estimates made concerning sleep stage can be substantially more accurate.
The method may be a computer-implemented method.
The method may be performed on a bed or seat where the sleeping person is located.
The input may include a rocking or reciprocating motion.
The method may be performed, at least in part, using control data generated by a machine learning model according to the third aspect of the disclosure for estimating one or more sleep stage parameters.
According to a second aspect of the present disclosure, there is provided a system for improving sleep quality for a sleeping person, the system comprising:
The advantages are similar to those discussed for the first aspect of the disclosure.
The system may be configured to perform the method of the first aspect of the disclosure. The system may include any feature or features presented with respect to the first aspect of the disclosure.
According to a third aspect of the present disclosure, there is provided a method of generating, by a machine learning model, control data for controlling a sleep enhancing device or system for improving sleep quality, the method comprising the steps of:
This aspect of the disclosure generates control data which are customized or tailored to a particular person's sleep pattern or sleep stages, for controlling a sleep enhancing device/system in an optimum way for that particular person. This can substantially improve their overall sleep quality by controlling operation of a sleep enhancing device/system in a way that complements the sleep stage that the person is in at various points during a sleep. Other advantages are similar to those discussed for the first aspect of the disclosure.
The processing step may include estimating, by the trained machine learning model, when the sensor readings indicate that the person enters one of sleep stage 3 (N3), sleep stage 4 (REM) or another sleep stage. The control data may include instructions to, when the person is estimated to have entered sleep stage 3 (N3) or sleep stage 4 (REM), stop or adjust the input provided by the sleep enhancing device or system.
This means that when the user is in a deep sleep, the movement of the bed can be reduced to a minimum or stopped. This can reduce possible adverse effects resulting from the sleep enhancing device or system, such as adverse effects from an oscillatory/rocking movement.
The processing step may include estimating, by the trained machine learning model, when the sensor readings indicate that sleep stage 3 (N3) and/or sleep stage 4 (REM) has elapsed by a predetermined extent. The control data may include instructions to, when the relevant sleep stage is estimated to have elapsed by the relevant predetermined extent, start or adjust the input provided by the sleep enhancing device or system.
For example, the instructions may lead to bed rocking at a predetermined speed or may increase the bed rocking speed. This can help to prolong sleep stage 3 and/or end sleep stage 4, relative to what would otherwise happen for the person during non-augmented sleep.
The one or more other sleep stages may include sleep stage 1 (N1) and/or sleep stage 2 (N2). The control data generated may be based on the estimates of when the sensor readings indicate that the person is in any one of sleep stage 3 (N3), sleep stage 4 (REM), and sleep stage 1 (N1) and/or sleep stage 2 (N2).
The control data may include instructions for controlling a rocking or reciprocating motion of the sleep enhancing device or system, including one or more bed oscillatory or reciprocation speed values.
The instructions in the control data may correspond to various features of the first aspect of the disclosure.
The sensor readings may be acquired when the person was or is asleep in an oscillating or reciprocating bed.
The data set may include a plurality of bed oscillatory or reciprocation speed values corresponding to each one of the sensor readings at the time that each one of the respective sensor readings was taken.
This allows data from a rocking bed to be taken into account when determining how a person's sleep stages change during sleep in a rocking bed as opposed to a conventional non-rocking bed.
The sensor readings may be from any one, two or more, three or more, four or more, five or more, or six or more of the following types of sensor: motion sensor, body temperature sensor, heart rate or pulse sensor, blood pressure sensor, EEG sensor, breathing rate sensor, ECG sensor, heart rate variability sensor.
As discussed earlier, using data from multiple sensors allows for greater accuracy when estimating which sleep stage a person is in at various points during their sleep.
If training the machine learning model ab initio, prior to processing a specific person's sleep data, then this may involve any one, some or all of:
Note that supervised learning occurs by feeding previously captured data from sensors and the desired output bed movement control data to the algorithm, with the aim of training the model to yield the desired output; whilst unsupervised learning through which the algorithms learn from scratch and decide which is the best solution to control the bed movement, given a subject specific sleep peculiarities that might not fit into the previously known patterns or clusters.
According to a fourth aspect of the present disclosure, there is provided a non-transitory computer readable medium containing program instructions for causing a processing device to perform a method of controlling a sleep enhancing device or system for improving sleep quality, the instructions including:
The advantages are similar to the first and second aspects of the disclosure.
According to a fifth aspect of the present disclosure, there is provided a non-transitory computer readable medium containing program instructions for causing a processing device to perform a method of generating, by a machine learning model, control data for controlling a sleep enhancing device or system for improving sleep quality, the instructions including:
The advantages are similar to the fourth aspect of the disclosure.
A sleep enhancing device or system (such as a bed rocking device or a rocking chair or a motion imparting leg(s) for a bed) may be provided which comprises the non-transitory computer readable medium of the fourth or fifth aspects of the disclosure.
Any aspect of the disclosure may include any feature or independently selected combination of features presented with respect to any other aspect of the disclosure.
Many aspects of the disclosure can be better understood with reference to the following drawings. The components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present disclosure. Moreover, in the drawings, like reference numerals designate corresponding parts throughout the several views.
Referring first to
In this embodiment, four bed legs 11 are provided under the four corners of the base. The bed legs 11 are motion imparting legs that can impart a reciprocating lateral rocking motion to the bed 10 so that it rocks back and forth in a horizontal plane when the legs are operating. Examples of the type of leg that can achieve this are presented in the Applicant's patent application WO2018083153 or the Applicant's patent application EP4137008, the disclosures of each of which are incorporated by reference.
The legs 11 can be interconnected (e.g. wirelessly) and preferably operate in a synchronized manner, so that the legs 11 at the left side of the bed 10 work antagonistically with the legs 11 on the right side of the bed 10 in rocking the bed. During use, when the legs 11 are turned on, the bed 10 can be displaced laterally left and right in a reciprocating manner within the limits of displacement provided by the legs.
The bed arrangement depicted has five different types of sensor device for sensing various parameters related to the person 12, for use in estimating which stage of sleep they are in at a given time during a sleeping episode. The sensor devices in this embodiment include: a wristwatch or bracelet 14, a chest belt 16, a headband 18, a mat sensor 20, and an accelerometer 22. It will be appreciated that the form of each type of device is not limited only to the depicted versions.
The bracelet or wristwatch 14 can have sensors for any of motion, temperature, measure heart rate pulse, electrocardiogram (ECG), and blood pressure. The chest belt 16 can have sensors for any of movement, temperature, heart rate/pulse measurement, electrocardiogram (ECG), and heart rate variability (which can be derived from ECG). The head band 18 can have sensors for any of motion, temperature, and EEG (electro encephalogram). The mat-type sensor 20 can have sensors for any of motion, breathing rate, heart rate/pulse, and heart rate variability.
It will be appreciated that any one, two, three or four, or all five, sensor types of the sensor devices may be used in any embodiment, but that it is preferred to use multiple sensors in some cases. It will also be appreciated that some of the sensor devices could be provided in a combined device, where a given sensor device has sensors common to two of the different ones illustrated. Examples of other sensor devices include a smartphone and a pulse oximeter.
An arrangement of this type can be used for a number of different people to record sleep data covering a wide range of body parameters during a series of sleep cycles.
In respect of supervised training, the following tables show the mapping of the information that is derived from each type of sensor in relation to each of the sleep stages (N1, N2, N3, REM). However, the ECG and heart rate variability sensors currently require specialized analysis algorithms (optionally using AI) to interpret how the sensor reading corresponds to sleep stage and so are not summarized below.
That is, in this case, the AI algorithm is told which sleep stage is attributed to each sensor reading for those sensors in the tables.
Note that there may be a step in which a doctor or a Smart AI software (which can analyze the sleep data e.g. from a whole night) provides or determines sleep stage information or decisions in respect of ECG and/or heart rate variability sensor readings.
Afterwards, the ECG and/or heart rate variability data and the corresponding sleep stage for each epoch (30 or 60 seconds) can be fed to the AI algorithm for supervised training.
After collecting sensor data from a suitable sample size of different people, the data can be used as part of training a machine learning model to estimate which sleep stage a sleeping person is in based on sensor readings from one or more sensors.
In the case of supervised learning, the machine learning model may be provided with the sensor data and corresponding sleep stage information (e.g. derived from the sensor data or determined by a doctor). The sensor data may be replaced or supplemented by polysomnogram data. This is advantageous because the sleep stage information derived from a polysomnogram is much more accurate, thus improving the training of the machine learning model.
In the case of unsupervised learning, the machine learning model may be trained without having sleep stage information inputted, so that the output of the model/AI algorithm is the determination or estimation of the sleep stage attributable to each sensor reading. However, this is currently less accurate than sleep stage information determined by other techniques noted above, and so supervised learning for the machine learning model can result in better outcomes.
In preparing the collected data for this purpose, the instantaneous sleep stage of the analyzed subject in each case can be accurately estimated by using known methods in somnology (i.e. by a medically trained professional/doctor, or by smart AI software-preferably utilizing polysomnogram data), in order to derive the sleep stage level starting from the specific and particular information provided by each type of sensor.
It will be appreciated that AI algorithms are able to undergo:
It is of course appreciated that the quality of a training data set is very important in order to accurately train an AI algorithm. Thus, as mentioned above, in some preferred embodiments it may be advantageous to obtain/use a polysomnogram for use in the case of supervised training, along with data from any of the above sensors. A polysomnogram very accurately captures the vital parameters of a person's body, and deriving sleep stage information based on the polysomnogram (and not from the other sensors above) can result in a more accurately trained model.
It will also be appreciated that some sensor readings/signals, such as EEG and ECG, could be pre-processed and/or transformed, for example into a frequency domain, with the aim of improving the accuracy of a model (once trained on the data). Automatic sleep stage classification by analyzing EEG and ECG signals may be beneficial as part of the process, if available.
In some embodiments, the sleep stage at which the various sensor readings were recorded could be determined automatically by a software program, or determined manually by a trained specialist (such as a doctor).
Once the sensor readings have been correctly categorized according to the sleep stage that they correspond to, the sensor readings and the sleep stage information could be fed to the AI algorithms for supervised training.
However, it can be beneficial to filter, clean up, window, down sample and/or normalize the data first. The data clean-up and normalization module indicated in the flowchart 100 has the role of translating/adapting all the sources of data from the various sensors and the bed oscillatory motion speed to:
For (a), EEG and ECG sensors may have a sampling frequency (i.e. rate of taking readings) in the range 200 Hz to 500 Hz. Sensors for motion, body temperature, heart rate/pulse, heart rate variability, breathing rate and/or bed oscillatory motion speed may have a sampling frequency of about 1 Hz.
For (b), EEG, ECG and/or motion sensor readings may be the range −1.0 to +1.0. Sensor readings for heart rate/pulse, heart rate variability and/or breathing rate may be in the range 0 to 256. Sensor readings for body temperature may be in the range 35 to 41 degrees Celsius or 95 to 106 degrees Fahrenheit. Sensor readings for bed oscillatory motion speed may be in the range 0 to 100 (which may correlate to: 0 to about 10 mm per second).
The data clean-up and normalization module may use:
This can result in cleaner dataset that can in turn lead to more accurate estimations/predictions from the model (once trained).
The data clean-up and module can also handle windowing of the incoming data into a standard length set (e.g. processing frame size). This may correspond to 30 seconds or 60 seconds, for example, although the window size value could be set to another value instead. In view of the large quantity of data accumulated over time by running an AI algorithm, this may reduce the system large memory requirements by down sampling data that carries redundant information.
The next module represents a person (e.g. supervisor or AI software engineer) who classifies data for supervised or unsupervised processing by setting appropriate flags. This is done for producing training, test, blind and run time data sets, which are respectively used for the creation, testing, verification/validation and real time run of the machine learning model/algorithm. Each of these is a well known step for building data sets for a machine learning model.
The control rules for the classification module are:
The control flags for the classification module are:
Each of the modules immediately after the classification module is a well-known type of AI algorithm which can predict with a certain confidence and satisfies the requirements for training a machine learning model on the sleep data set: clustering, case-based reasoning, data mining, LSTM (Long Short-Term Memory) neural network.
The clustering module is capable of user sleep classification into clusters which cover sleep disorders/dysfunctions and normal types of sleep behavior.
The case-based reasoning module is capable of addressing new problems by retrieving stored ‘cases’ describing similar prior problem-solving episodes and adapting the solutions to fit the new needs.
The data mining module is capable of extracting, deriving or abstracting rules from large quantities of data.
The LSTM neural network is capable of long-term and short-term sleep stage prediction.
A decision engine then needs to select the most relevant information from each of the four above AI modules. Along with user set control rules and flags, it then produces the sleep stage confidence levels/thresholds and the flags for when to start, stop or adjust the input (e.g. bed motion) in sleep stage 3 and/or sleep stage 4, and optionally sleep stage 1 and/or sleep stage 2.
Note that in sleep stage 1 and/or sleep stage 2, the input may be a user-selected preset which has a non-zero value. For example, for a sleep enhancing device including a motion imparting device for reciprocating rocking of the bed, a non-zero speed for bed reciprocation may be used.
Typically, the flag or trigger upon which to initiate or “switch on” motion for sleep stage 3 will be set at a few minutes before the predicted lapsing time of sleep stage 3. The exact point at which this is set is based on the deemed success rate of previous starts. That is, if in a previous sleep stage 3 the input was set at time t−5 mins from the predicted end time t of a sleep stage 3, but the user in fact transitioned out of the sleep stage 3 earlier than predicted, then in a subsequent sleep stage 3 the input will be scheduled to start suitably earlier to mitigate the likelihood of starting too late again.
Typically, the flag or trigger upon which to initiate or “switch on” motion for sleep stage 4 will be set at about half of the usual time that is spent in sleep stage 4, i.e. at around the midway point of sleep stage 4. The REM period duration is different for each sleep cycle, and increases throughout the night. For example, there is a shorter REM sleep period for the first sleep cycle and relatively longer REM sleep period for the later sleep cycles. This means that the length of time for which bed leg motion is ceased for sleep stage 4 may need to be adjusted over several nights, so that the AI algorithm can learn the person's specific REM sleep duration for each of their sleep cycles.
The control rules for the decision engine are:
The control flags for the decision engine are:
Having trained a machine learning model broadly as indicated above, or otherwise having access to a suitably trained model, a sleep data set acquired using some or all of the sensors or sensor devices (discussed with respect to
The AI model can then, according to its training, take the sensor readings from the available sensors and assess the likelihood that the person 12 is/was in a state corresponding to sleep stage 1 (N1), sleep stage 2 (N2), sleep stage 3 (N3), or sleep stage 4 (REM), and can provide a confidence stage/level for each of these. The confidence levels for all four sleep stages (that is, the probability that a particular sleep stage is the ‘active’ sleep stage at a given time) must always sum to 100%. It should be noted that only one sleep stage may be active at a time, but there might be edge cases when two of them are assigned equally large confidence levels.
The AI model can also, according to its training, take the sensor readings from the available sensors and identify markers or indications in the data that signal the onset and end of the various sleep stages, and in particular: markers or indications in the data that signal the onset of sleep stage 3, markers or indications in the data that signal the onset of sleep stage 4, markers or indications in the data that signal the end of sleep stage 3, and markers or indications in the data that signal the end of sleep stage 4. This allows for estimation of the length of each of that person's sleep stages, accounting for variability in the length of the sleep stages as sleep cycles progress.
Control data can be generated by the trained machine learning model based on the data corresponding to the person 12. The control data include instructions concerning when to start an input and/or when to stop an input and/or when to adjust (increase/decrease/change) an input, which in this case is a motion speed imparted by the bed legs 11.
In particular, the control data can start/stop/change an input at the onset of sleep stage 3, at the onset of sleep stage 4, near the end of sleep stage 3, and/or around the middle of sleep stage 4, although other points may additionally or alternatively be selected for taking action regarding the input. Example criteria for starting/stopping/changing the input are discussed above with respect to
The control data thus prepared can be used to control the input provided by the bed legs 11 each time that the person 12 sleeps on the bed 10, so that the bed legs 11 rock the bed 10 in a way that improves that person's sleep cycles and relative abundance of sleep stages within those cycles. In particular, the control data is intended to improve the duration of sleep stage 3 (N3) for the person when they sleep and to relatively controllably alternate between sleep stage 3 (N3) and sleep stage 4 (REM) for the person 12 when they sleep.
It will be appreciated that, in implementing the present disclosure, the algorithm(s) could run locally on a processor of the rocking bed 10 (or the legs 11 thereof), or on a server located on the internet (e.g. in the cloud) which could include bidirectional data communication and control between the moving bed 10 and the cloud.
If using a cloud configuration, captured real-time data from sensors could be sent to the cloud optionally along with the bed's instantaneous oscillatory movement parameters (e.g. speed), and in the other direction real-time commands/instructions with the bed's motion parameters (e.g. bed oscillatory motion speed information) can be sent from the cloud to the moving bed 10.
It is possible for the sensors to be connected directly to the rocking bed (e.g. to the processor) or the sensors to communicate through other methods (e.g. the internet) with the cloud in which the artificial intelligence algorithms could run.
A hardware platform that support algorithms for neural network computation may be used to implement the disclosure. The functional blocks of the method may be independently operated on a microservice platform.
It will be appreciated that the disclosure has potential applications to provide better sleep in various different scenarios. This includes scenarios beyond sleeping in a bed. For example, implementing the disclosure in a seat in a vehicle such a plane or train or boat may help improve sleep quality for a passenger.
The embodiments described above are provided by way of example only, and various changes and modifications will be apparent to persons skilled in the art without departing from the scope of the present disclosure as defined by the appended claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2400080.4 | Jan 2024 | GB | national |
