Wearables may be a $4.5 billion dollar market, but the devices themselves can be, well, annoying to wear: besides being a constant presence against your skin, you also have to remember to put them on in the first place. What’s more, if you’re one of millions Americans who have a sleep disorder or just experience difficulty sleeping, wearables don’t get the medical-grade data doctors need. Instead, you have to go to a sleep lab and spend the night with electrodes taped to your skull. When it comes to sleep-monitoring technologies, these are significant barriers to a positive user experience–and widespread adoption.
Now, a group of researchers at MIT have developed a remote sleep sensing system that uses radio waves to capture data about your brain waves while you sleep–and AI to read them–without ever touching your body. It consists of a laptop-sized wireless device that emits radio signals. When put in the user’s bedroom, the waves detect even the slightest movement of the body. The system doesn’t just do the job of a sleep-tracking wearable without the wearable; it also just provides data at a similar level of accuracy as a sleep lab.
“You don’t have to do anything–that’s the beauty of it,” says Dina Katabi, a professor of electrical engineering and computer science at MIT. “You hang it on the wall and you forget about it and it just works and gives you the data.”
Katabi explains that the types of low-frequency radio waves that her device uses often aren’t very accurate. It’s one thing to point radar at the open sky to detect an airplane, and quite another to use waves within a small, enclosed space to detect subtle physiological signs like breathing and heart rate.
In order to cut out all the extraneous information her system records, she developed a machine learning algorithm that can extract sleep stages–light, deep, and REM sleep–out of the mess of data. The algorithm was trained on a sleep dataset of 25 individuals for a total of 100 nights of sleep, taken using an FDA-approved device that uses EEG to record brain waves. For those same 25 participants and 100 days, Katabi’s experimental device recorded data as well. The EEG recordings provided the scientific data needed to train the more experimental system, helping it learn what data was extraneous and what was relevant to identifying sleep stages.
The results are 80% accurate–which doesn’t sound very good at first. But Katabi explains that reading sleep data is a subjective skill often done by sleep specialists. She says that two sleep specialists’ readings of one night of sleep data can be different, and that 80% accuracy is on par with what you might expect going to a sleep lab. Other researchers’ remote sleep systems have an accuracy of 65%, but can only determine if someone is awake or asleep, not their sleep stage.
Unfortunately, you won’t be able to buy one of these devices anytime soon. Katabi wants to focus on using the sleep monitoring device to study disorders that impact sleep, like Parkinson’s, depression, and Alzheimer’s. She’s already partnering with doctors at Massachusetts General Hospital who can use it to monitor their patients.
Because the device produces medical-grade data, it’s difficult to compare to wearables already on the market which typically use accelerometers to track sleep. “[Wearables are] good, it’s better than nothing, but it’s really not enough to provide something that would be at the level of accuracy for the doctor,” Katabi says. But wearables provide a useful point of comparison as to why her device’s ability to remotely track sleep is so important, especially in a medical context. “A large part of the population are not really that interested in wearables,” she says. “We’re talking about Alzheimer’s patients. They’re not going to remember to put the wearable on.”
Instead, Katabi refers to her technology as something else entirely. “So we talked about wearables,” she says. “We call these kinds of technologies ‘the invisibles.'”