Home » Health »

New AI algorithm tracks sleep using radio waves

 

Scientists have finally discovered a non-intrusive way to study sleep that doesn’t rely on bothersome sensors.

A new algorithm developed by researchers at MIT and Massachusetts General Hospital makes is possible to track a patient’s sleep wirelessly over radio waves.  

It lives in a device – about the size of a laptop – that simply has to be near the user to analyze the radio signals around the person and translate those measurements in order to track light, deep and rapid eye movement (REM) sleep stages.

Scroll down for video 

The new algorithm lives in a device - about the size of a laptop - that simply has to be near the user to analyze the radio signals around the person and translate those measurements in order to track light, deep and rapid eye movement (REM) sleep stages

The new algorithm lives in a device - about the size of a laptop - that simply has to be near the user to analyze the radio signals around the person and translate those measurements in order to track light, deep and rapid eye movement (REM) sleep stages

The new algorithm lives in a device – about the size of a laptop – that simply has to be near the user to analyze the radio signals around the person and translate those measurements in order to track light, deep and rapid eye movement (REM) sleep stages

HOW IT WORKS 

Users simply place the device near the bed, such as on a wall.

While the person is sleeping, the sensors in the wireless device emit low-power radio frequency (RF) signals.

As they reflect off the body, any slight movement alters the frequency of the reflected waves. 

Then all that’s left to do is analyze the waves by translating their measurements of pulse, breathing rate, and movement into sleep stages.

A new and advanced deep neural network is able to do so by separating the useful information needed to determine sleep stages from all the other irreverent information collected.

The RF waves are replacing the need for attached sensors, and the algorithm is analyzing the data just like a EEG lab technician would. 

More than 50 million Americans alone suffer from sleep disorders and other conditions that can affect sleep. 

Until now, there has been no good way for doctors to monitor and diagnose sleep problems – the only methods available relied on attaching electrodes and a variety of sensors to the patient (usually in a lab), which can disrupt sleep even further and inhibit doctors’ ability to properly understand the issue.

The new artificially intelligent algorithm eliminates that problem completely.

‘Imagine if your wifi router knows when you are dreaming, and can monitor whether you are having enough deep sleep, which is necessary for memory consolidation,’ says Dina Katabi, the Andrew and Erna Viterbi Professor of Electrical Engineering and Computer Science, who led the study.

‘Our vision is developing health sensors that will disappear into the background and capture physiological signals and important health metrics, without asking the user to change her behavior in any way.’ 

Until now, there has been no good way for doctors to monitor and diagnose sleep problems - the only methods available relied on attaching electrodes and sensors to the patient (usually in a lab), which can disrupt sleep  further and inhibit doctors' ability to understand the issue

Until now, there has been no good way for doctors to monitor and diagnose sleep problems - the only methods available relied on attaching electrodes and sensors to the patient (usually in a lab), which can disrupt sleep  further and inhibit doctors' ability to understand the issue

Until now, there has been no good way for doctors to monitor and diagnose sleep problems – the only methods available relied on attaching electrodes and sensors to the patient (usually in a lab), which can disrupt sleep  further and inhibit doctors’ ability to understand the issue

Katabi along with four other researchers who worked on the project presented their findings at the International Conference on Machine Learning on Aug. 9.

‘We introduce a new predictive model that combines convolutional and recurrent neural networks to extract sleep-specific subjectinvariant features from radio frequency (RF) signals and capture the temporal progression of sleep,’ reads the paper.

Previously, Kabati and other researchers at the school have done the same thing to measure other vital signs and behaviors like walking speed with a sensor called WiGait, but this is their first go at applying the technology to sleep.

Three examples of full night predictions corresponding to the average, best and worst classification accuracy

Three examples of full night predictions corresponding to the average, best and worst classification accuracy

Three examples of full night predictions corresponding to the average, best and worst classification accuracy

In more detail, the system works because the sensors in the wireless device emit low-power RF signals, and as they reflect off the body, any slight movement alters the frequency of the reflected waves.

Then all that’s left to do is analyze the waves by translating their measurements of pulse, breathing rate, and movement into sleep stages.

Using a new and advanced deep neural network, they’re able to separate the useful information needed to determine sleep stages from all the other irreverent information collected.

This was the true feat, because while the team already had the RF technology working to collect vitals, algorithms existing prior to the new one they created were unable to properly and effectively analyze the data.

 The sensors in the device emit low-power RF signals, and as they reflect off the body, any slight movement alters the frequency of the reflected waves. Then all that’s left to do is analyze by translating their measurements of pulse, breathing rate, and movement into sleep stages

‘The surrounding conditions introduce a lot of unwanted variation in what you measure,’ says Thomas Siebel, Professor of Electrical Engineering and Computer Science and a member of the Institute for Data, Systems, and Society at MIT, who also worked on the project.

‘The novelty lies in preserving the sleep signal while removing the rest.’ 

The purposely trained the algorithm to ignore wireless signals that bounce off of other objects in the room and include only data reflected from the sleeping person. 

Additionally, their algorithm can be used in different locations and with different people, without any calibration. 

Figure 2. 2 (a) shows how the model can distinguish deep and light sleep with high accuracy. And 2(b) illustrates that the model works well for different subjects and environments

Figure 2. 2 (a) shows how the model can distinguish deep and light sleep with high accuracy. And 2(b) illustrates that the model works well for different subjects and environments

Figure 2. 2 (a) shows how the model can distinguish deep and light sleep with high accuracy. And 2(b) illustrates that the model works well for different subjects and environments

In the first trial, the researchers tested the method on 25 healthy volunteers.

They had 80 percent accuracy, which is comparable to the accuracy of ratings determined by sleep specialists based on intrusive EEG sleep sessions in labs.

‘Our device allows you not only to remove all of these sensors that you put on the person, and make it a much better experience that can be done at home, it also makes the job of the doctor and the sleep technologist much easier,’ Katabi says. 

‘They don’t have to go through the data and manually label it.’ 

This is not the first time researchers have attempted to sue RF to study sleep, but it’s the most successful attempt yet. 

Other systems could only determine if a person is asleep or awake but not which stage of sleep they’re in – additionally, they only saw 65 percent success. 

 Katabi along with four other researchers who worked on the project presented their findings at the International Conference on Machine Learning on Aug. 9

Going forward the researchers hope to apply the technology to sleep disorders such as insomnia and sleep apnea as well as other conditions like Parkinson’s disease, Alzheimer’s and seizures that occur during sleep.

‘When you think about Parkinson’s, you think about it as a movement disorder, but the disease is also associated with very complex sleep deficiencies, which are not very well understood,’ Katabi says. 

Amazon covertly opened a secret lab called '1492' that's dedicated to overhauling healthcare. Headquartered in Seattle, Washington, the skunkworks facility is focused on both hardware and software projects and is looking into 'telemedicine'

Amazon covertly opened a secret lab called '1492' that's dedicated to overhauling healthcare. Headquartered in Seattle, Washington, the skunkworks facility is focused on both hardware and software projects and is looking into 'telemedicine'

Amazon covertly opened a secret lab called ‘1492’ that’s dedicated to overhauling healthcare. Headquartered in Seattle, Washington, the skunkworks facility is focused on both hardware and software projects and is looking into ‘telemedicine’

SECRET ‘1492’ JOBS

Even the online job posting for the lab are hush-hush.

 On Amazon’s job board, the secret roles are listed as ‘a1.492’ and described as being for ‘The Amazon Grand Challenge a.k.a. ‘Special Projects’ team.’

On LinkedIn, some employees on the project listed their affiliation with ‘a1.492.’

This includes multiple machine learning experts, which leads to the belief that Amazon’s healthcare approach will be AI-heavy.

As of late, sleep has been a major topic of interest for researchers and tech giants alike.

In late July, reports claimed Amazon covertly opened a secret lab called ‘1492’ that’s dedicated to overhauling healthcare.

Headquartered in Seattle, Washington, the skunkworks facility is focused on both hardware and software projects, and is looking into ‘telemedicine.’

Anonymous sources close to Amazon told CNBC the company has become ‘increasingly interested in exploring new business in healthcare.’

Amazon already has a unit looking into selling pharmaceuticals, and now this team is looking to push and pull data from ‘legacy’ medical electronic systems and build a platform for ‘telemedicine’ that would involve virtual consultations with doctors.

While it’s not known if Amazon is looking to build any new health devices, the sources did not rule it out.

The secret team is, however, already known to be exploring health applications for existing Amazon devices like the Dash Wand and Echo, the latter of which already has health-focused ‘skills’ – or voice commands – developed by doctors and hospitals.

Apple's new system will monitor how well you sleep and then send sleep notifications each evening

Apple's new system will monitor how well you sleep and then send sleep notifications each evening

The system will send sleep summaries based on the data it collects during the week. It will tell users how long it took them to get to sleep (SOL) and the number of hours they slept each night

The system will send sleep summaries based on the data it collects during the week. It will tell users how long it took them to get to sleep (SOL) and the number of hours they slept each night

The system will send sleep summaries (right) based on the data it collects during the week. It will tell users how long it took them to get to sleep (SOL) and the number of hours they slept each night. It will also send sleep notifications (left) each evening

HOW IT WORKS 

The patent details a system that uses Apple mobile devices like the Apple Watch or iPhone.

These devices will use a heartbeat sensor, microphone, breathing rate sensor, movement sensor and more to track when users are asleep.

It will use this data to send them regular sleep reminders and sleep quality reports.

It will even set your alarm back if you took too long to get to sleep or woke up during the night. 

Earlier this summer, Apple took another leap into the sleep tracking world as well.

In June, the firm was granted a patent this week that details the use of devices like the iPhone and Apple Watch to track sleep as Apple expands its work into sleep science.

The patent details automatically ‘adjusting alarms based on sleep onset latency’ using sleep data collected by mobile devices. 

Using an Apple Watch or iPhone’s heartbeat sensor, microphone, breathing rate sensor, movement sensor and more, the technology will know when users are asleep.

It will send users a regular sleep quality report with a rundown of how many hours rest they have had.

If users have woken up in the night or gone to bed late, the system will automatically push their alarm back to give them more sleep.

The system will also send users regular reminders to get their sleep schedule in order if their routine becomes erratic.

The technology could be implemented with iOS 11 or perhaps future versions of iOS.

The system will give users the option to set normal alarms, but it could push that time back if it determines you aren't getting enough sleep

The system will give users the option to set normal alarms, but it could push that time back if it determines you aren't getting enough sleep

Using a heartbeat sensor, microphone, breathing rate sensor, movement sensor and more on the Apple Watch or iPhone, the technology will know when users are asleep. Users will have the option to set timed naps (pictured)

Using a heartbeat sensor, microphone, breathing rate sensor, movement sensor and more on the Apple Watch or iPhone, the technology will know when users are asleep. Users will have the option to set timed naps (pictured)

Using an Apple Watch or iPhone’s heartbeat sensor, microphone, breathing rate sensor, movement sensor and more, the technology will know when users are asleep. They will have the option to set timed naps (right) as well as normal alarms if needed (left)

‘In some implementations, a mobile device can adjust an alarm setting based on the sleep onset latency duration detected for a user of the mobile device,’ Apple writes in its patent.

‘For example, sleep onset latency can be the amount of time it takes for the user to fall asleep after the user attempts to go to sleep (e.g., goes to bed).

‘The mobile device can determine when the user intends or attempts to go to sleep based on detected sleep ritual activities.’

Apple says that these sleep ritual activities are things a user does before going to sleep, such as brushing their teeth.

The app (picturd) learns user's sleep patterns and connects the data to daily habits in order to help them get a better night's rest

The app (picturd) learns user's sleep patterns and connects the data to daily habits in order to help them get a better night's rest

Apple purchased the app and sleep monitoring device Beddit last month (pictured), which learns user’s sleep patterns and connects the data to daily habits in order to help them sleep

With the sleep industry estimated to be worth more than £25 billion ($32 billion), Apple has recently made a number of big moves into sleep science and technology (stock image)

With the sleep industry estimated to be worth more than £25 billion ($32 billion), Apple has recently made a number of big moves into sleep science and technology (stock image)

With the sleep industry estimated to be worth more than £25 billion ($32 billion), Apple has recently made a number of big moves into sleep science and technology (stock image)

‘In some implementations, the mobile device can determine recurring patterns of long or short sleep onset latency and present suggestions that might help the user sleep better or feel more rested.’ 

With the sleep industry estimated to be worth more than £25 billion ($32 billion), Apple has recently made a number of big moves into sleep science and technology.

Last month, the firm purchased an app and sleep monitoring device called Beddit, which learns users’ sleep patterns and connects the data to daily habits in order to help them get a better night’s rest.

It has been speculated that Apple could add some type of sleep tracking technology to the watchOS 4 that is set to be released later this year. 

 

Related Posts

  • No Related Posts