Massachusetts General Hospital and MIT researchers have developed a new method to monitor sleep stages without any type of sensor attached to the body. The device runs on an advanced AI algorithm, and captures and analyzes the radio signal around the human body. It further translates those data into sleep stages – deep, light or REM (Rapid Eye Movement).
The aim is to develop health sensors that work in the background and capture crucial health metrics and physiological signals, without asking you to alter your behavior in any way.
It’s like a Wi-Fi router that knows when you are sleeping and dreaming, and can tell you whether you are having enough deep sleep, which is by the way essential for memory consolidation.
Dina Katabi, professor at MIT, has previously designed radio-based sensors, which remotely analyzes vital behaviors and signals that could be indicators of health. These sensors contains a wireless device (similar to laptops) that emits low-power RF (Radio frequency) wave.
Since the radio waves reflect human body in various manners, any small movement of the body changes the radio frequency. Analyzing these reflected waves could reveal important signals like breathing rate and pulse.
Katabi used the same approach to develop WiGait – a sensor that can measure walking speed via radio wave. This could help physician anticipate cognitive falls, particular pulmonary disease or cardiac, or other health issues.
After building these kinds of sensors, Katabi came up with a similar idea – develop a wireless device for monitoring sleep. This would help more than 50 million Americans who are suffering from sleep disorders and similar illness.
How It Works?
Image credit: YouTube
To monitor sleep wirelessly, scientists came up with a new way to translate measurement of breathing rate, pulse and movement into sleep phases. With the help of advanced artificial intelligence algorithms like deep neural network, it is possible to extract and analyze complex datasets, which in this case is radio signals captured by sensors.
These signals also contain a lot of data unrelated to sleep, and can be ambiguous or confusing to AI algorithms. Therefore, scientists developed a new artificial intelligence algorithm based on deep neural network to get rid of these irrelevant data.
One more factor that introduces unwanted/irrelevant variation is surrounding conditions. For better efficiency the algorithm is used in various locations with different subjects, without any calibration.
Tests and Results
This new methodology is tested on 25 healthy volunteers, and scientists discovered that their approach was around 80% accurate as compared to the accuracy of Electroencephalography measurements.
It is true the many scientists have used radio frequency to monitor sleep, but their systems were returning false results 35% of the time. Also, they weren’t able to determine what sleep stage a person is in; they could only tell whether a subject is asleep or awake.
The MIT scientists are now planning to use this technique to learn how Parkinson’s disease (a progressive movement disorder) affects sleep. The sensors will also help them to study other sleep disorders like sleep apnea and insomnia. It may also be useful for studying sleep changes generated by Alzheimer’s disease (most common form of dementia) and epileptic seizures, which occur when you sleep.