Sensors Can Identify Sleep Stages
By Mike Bassett
A new device utilizes an advanced artificial intelligence algorithm to analyze radio signals for analysis of older adults' sleep patterns and translate data into sleep stages.
Getting a good night's sleep can be a real problem for older adults. In fact, according to the American Sleep Association, 50 to 70 million American adults have sleep disorders—conditions that have been associated with a wide range of adverse consequences, including an increased risk of hypertension, diabetes, obesity, heart attack, and stroke.
And significant changes in sleep, the causes of which can include medical illnesses, depression and other psychiatric disorders, and sleep disorders such as obstructive sleep apnea or restless leg syndrome, often accompany aging for older adults. In many of these cases a diagnosis requires a sleep study, but the standard manner in which sleep studies are conducted is inconvenient and intrusive.
The gold standard for monitoring sleep is polysomnography (PSG). This is typically conducted in a sleep lab or hospital and requires a patient to be connected to a large number of sensors, such as EEG-scalp electrodes, ECG monitors, chest bands, and nasal probes.
As a result, patients undergoing such studies can experience sleeping difficulties, which can affect their results. In addition, the costs involving in running such studies, and often the discomfort involved in participation, limit the potential for long-term sleep studies.
However, researchers at the Massachusetts Institute of Technology (MIT) and Massachusetts General Hospital led by Dina Katabi, PhD, the Andrew & Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT, have developed a device that uses an advanced artificial intelligence algorithm to analyze radio signals around a subject and translate the measures into sleep stages.
How It Works
For example, Katabi and her colleagues from MIT, Hong Kong University of Science and Technology, and the Boston University School of Medicine, have developed a device called WiGait that can wirelessly measure walking speed with the idea this could help physicians monitor and diagnose health issues such as dementia and cardiovascular disease that are particularly relevant to elderly individuals.
Using these sensors, Katabi and her colleagues wanted to determine whether a similar approach could be used to monitor sleep. However, there are challenges involved in using RF measurements for sleep staging, especially the fact that RF signals carry large amounts of information that is irrelevant to sleep staging and dependent on the individuals involved and environmental factors.
What Katabi and her colleagues did was develop an artificial intelligence algorithm that preserves sleep signals while removing irrelevant information so that it can be used on people in different locations without having to be recalibrated.
"Our device transmits a very low-power wireless signal that reflects off a human body," Katabi explains. "These reflections capture motion and change with our breathing and heartbeat. We have designed a deep neural networks architecture that analyzes these reflections and predicts stages of sleep. While standard sleep studies require the person to be hooked up with many sensors and electrodes, our solution doesn't put any sensor on the body, allowing the person to sleep comfortably in his own bed."
In a study carried out in collaboration with Matt Bianchi, MD, PhD, chief of the division of sleep medicine at Massachusetts General Hospital (MGH), MIT and MGH researchers tested the device on 25 healthy adults who used it over 100 nights of sleep.
"This was a validation study to see whether the device can be set up in the real world," Bianchi says. "Can it monitor someone in their natural surroundings over multiple nights and actually show whether the algorithm shows us what the brain was doing? And can we use just those two signals—breathing patterns and heart rates—detected by radar to tell us what the brain was doing whether asleep, awake, REM or non-REM?"
The researchers found that the device was able to detect sleep patterns with an accuracy rate of about 80%—comparable to that of PSG.
According to Katabi, a device that enables monitoring of sleep stages without the intrusiveness of sensors has a number of implications for sleep disorder studies. For example, because the device eliminates the need for uncomfortable on-body sensors that can affect the results of traditional sleep studies, it can help researchers better estimate the parameters needed to understand sleep disorders.
Additionally, "our system also makes long-term monitoring of sleep possible," Katabi says. "Today longitudinal sleep studies are very difficult, as it is hard to expect the person to sleep more than a few days in a sleep lab. In contrast, our device enables such studies in one's home and for months or years without overhead for the patient," she says.
The device could also be used as a medical grade device for the clinic, or even as a device geared toward health care consumers. "Our system does not need to be calibrated to each person," Katabi says. "The device can work with a new person whom it did not see before, so it will be easy to use it as a consumer device too."
Bianchi noted that the sleep industry is trending toward technology that is less cumbersome and intrusive. "Sleep apnea is fundamentally a breathing problem, so if a device measures breathing, we've already performed a lot of algorithm work showing that the breathing pattern of a patient can tell you something about their sleep apnea status," Bianchi says. "This may sound obvious, but it's not how we normally do this."
Instead, a sleep study usually requires five different sensors just to analyze sleep apnea, Bianchi says. "Its an important step for us to say we just [wirelessly monitored] the movement of the thorax instead of having an airflow monitor, or an oxygen sensor, or other muscle sensors like we would in a sleep lab. So I see the whole field doing more of this clinical diagnostic monitoring using fewer and fewer sensors, which makes something like this radar device more attractive for that reason."
Bianchi points out that it will likely take some time to determine exactly what kind of role this device will play when it is finally commercialized. For example, he notes that in 2007 the American Academy of Sleep Medicine released practice guidelines related to the standard of care for using limited channel home-testing kits for sleep apnea, but that Medicare and insurance coverage of that type of testing has lagged behind.
"And that first announcement in 2007 came after about 10 years of research on those devices," Bianchi says. "That's kind of a glacial time frame for clinical care to change. "This device will definitely have a role in clinical practice," he adds, "but translating the newest device findings into clinical practice takes time."
As for its relevance to elderly patients, Bianchi notes that this was a feasibility study involving healthy adults, and "what we can accomplish in a healthy adult might require a totally different set of algorithms for an elderly adult who may have a sleep disorder such as sleep apnea, may wander out of bed, or have other issues."
However, Katabi points out that the device could be used to understand many diseases of interest to the geriatric population, including depression, Parkinson's disease, and Alzheimer's disease, all of which cause sleep disturbance. "Furthermore," she adds, "our system is also very easy to use by older people or those who are not computer savvy because it does not require the user to wear sensors on the body, push buttons, or do anything special."
— Mike Bassett is a freelance writer based in Holliston, Massachusetts.