MIT Study Uses AI to Monitor Sleep Patterns

“Imagine if your Wi-Fi router knows when you are dreaming, and can monitor whether you are having enough deep sleep, which is necessary for memory consolidation,” posed Dina Katabi, lead author on a new study out of the Massachusetts Institute of Technology. Katabi, the Andrew and Erna Viterbi Professor of Electrical Engineering and Computer Science, was commenting on a new method developed by MIT and Massachussetts General Hospital that enables researchers to monitor sleep patterns without needing to attach electrodes to a patient.

“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,” she said in an MIT press release.

The device the team developed is, according to the same release, roughly the size of a laptop. It emits low-power radio frequency (RF) waves that bounce off of the sleeping subject, with any minor movement altering the frequency of their reflection. Their goal was to determine depth and quality of sleep using tells like pulse and breathing rate.

Artificial intelligence is at the core of the group’s ability to analyze those altered radio signals and develop an understanding of what precisely is happening in the sleeping subject. Using deep neural networks to eliminate unwanted variations irrelevant to the signals they were looking for, the team developed a novel AI algorithm to produce only the information they sought.

Katabi says the algorithm makes monitoring easier for the patient, because they don’t need to go to sleep covered in sensors, but also for the doctor and sleep technologist because, “They don’t have to go through the data and manually label it.”

The technique was tested on 25 healthy volunteers, and found to be “about 80 percent accurate, which is comparable to the accuracy of ratings determined by sleep specialists based on EEG [electroencephalography] measurements.” EEG systems typically consist of dozens of electrodes spanning across the top of a subject’s head.

The work is important, the team believes, because of the prevalence of sleep disorders: the Centers for Disease Control and Prevention estimates that between 50 and 70 million American adults suffer from some form of one. Other conditions, like Alzheimer’s and Parkinson’s, can also interfere with sleep. The MIT/MGH team plans to use the device and algorithm to next study the impact Parkinson’s has on sleep, which Katabi said is currently not well understood.

“We have this technology that, if we can make it work, can move us from a world where we do sleep studies once every few months in the sleep lab to continuous sleep studies in the home,” another of the study’s authors, Mingmin Zhao, exuded.

The team will present their full paper, “Learning Sleep Stages from Radio Signals: A Conditional Adversarial Architecture,” this week at the International Conference on Machine Learning in Sydney, Australia.

Related Coverage:MIT Team Tests Object-Identifying Algorithm for the Visually ImpairedMicrosoft, Machine Learning, and Better Mosquito DefenseAn Algorithm for Flu Outbreak Detection