News: 1615382225

A smart-home speaker device and machine-learning software can be used to pick up a person's heartbeat, and report the results to doctors to analyze, folks at University of Washington reckon.

Their paper, [1]published in Communications Biology, describes a “proof-of-concept system for acquiring individual heart beats using smart speakers in a fully contact-free manner.”

[2]

First, the speaker emits a steady series of high frequency beeps measuring anywhere between 18 to 22 kHz for over a minute – that's at the upper end of the range of frequencies humans can detect. These sounds are directed at someone sitting up to two feet away, and the microphones in the custom-built gadget detect the sound waves reflected off the human body.

Next, machine-learning algorithms process the audio clips from the returned waves and convert this signal into the rhythm of a heartbeat. Shyam Gollakota, co-author of the paper and an associate professor at the Paul G. Allen School of Computer Science & Engineering at UW, explained to The Register that the system uses sonar to detect tiny changes in the motion of someone’s skin.

[3]

"You can see the pulse in areas like someone’s chest or neck," he said. "These sub-millimetre changes are detected in the reflected inaudible sound waves. The algorithms inspect this data and convert it into a heart rate reading in beats per minute.

"It’s a bit like how submarines send out sound pulses to see their environment. They look at the time of arrival of the reflected pulses to figure out if an obstacle is moving towards or away. There are bits of the body that move in relation to your heartbeat, and through the reflected sound signal we can measure if it’s beating faster or slower."

Think your smartwatch is good for warning of a heart attack? Turns out it's surprisingly easy to fool its AI [4]READ MORE

The US-based team tested their contact-less, smart speaker technique on 49 patients; about half of them had healthy heartbeats, and the other half had cardiac conditions and irregularities in their heart rhythms. Each patient’s heartbeat was first measured using electrocardiography and then measured again using the smart speaker gizmo to compare the readings.

“Compared to ECG devices, we averaged an error rate of only about one to two beats. Smart speakers could be useful for telemedicine platforms," Gollakota said.

"Imagine if you’re sitting in front of a smart speaker and your computer can send your heartbeat information to doctors, that’s pretty helpful. It can be time consuming and expensive for patients to keep going to the doctors just for check-ups."

The prototype device contains seven microphones; consumer-grade voice assistants typically have between two and seven, depending on the model.

The machine-learning algorithm learns how to calibrate the incoming sound signals bouncing off a person’s skin. “We use a gradient descent algorithm to solve this optimisation problem,” Gollakota said.

Unlike neural networks, the algorithm designed by the researchers requires no training data. The algorithm described in the paper is less artificial intelligence, and more like a traditional software algorithm, where the features have been explicitly handcrafted. “For example, we make it so that the algorithm ignores the reflected signals that are from our chests moving up and down from breathing,” he added.

Gollakata is also part of a startup spun off the university, called Sound Life Sciences. The biz is trying to use consumer hardware to monitor health signals, and will try to commercialize the prototype’s technology.

[5]

The researchers imagine that devices could be used to monitor things like sleep apnea. Gollakata believes that something like blood pressure might one day be measured contact-lessly, too. ®

Get our [6]Tech Resources



[1] https://www.nature.com/articles/s42003-021-01824-9#Sec2

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YEj7LaJfvcCmQxZVQCP81QAAAEA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YEj7LaJfvcCmQxZVQCP81QAAAEA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[4] https://www.theregister.com/2020/03/10/ai_models_electrocardiogram/

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YEj7LaJfvcCmQxZVQCP81QAAAEA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[6] https://whitepapers.theregister.com/