"Alexa, watch my heart" – First circuit monitoring system for smart speakers


Smart speaker systems such as lifeguards

Heart disease is by far the most common cause of death worldwide. In the case of a heart attack or cardiac arrest, the most important factor is how quickly affected people receive professional help so that permanent damage and, above all, death can be avoided. American developers have already introduced an artificial intelligence system for smart speakers, such as Google Home and Amazon Alexa, which detects acute heart failure based on sounds and automatically calls an ambulance.

Researchers at the University of Washington have developed artificial intelligence that detects a heart attack in people without any touch or without sensors attached. All you need to know is a smart speaker, like the one offered by Google or by Amazon. With the help of striking noises, which people give in case of a heart attack, the AI ​​must recognize if there is an acute emergency. The monitoring system then independently calls a rescue service to help. The researchers recently presented their findings in the journal "npj Digital Medicine".

With a new AI, Alexa, Google home and CO. Recognize breathing if there is a cardiovascular emergency. (Image / reading / fotolia.com)

OK Google: Examine my heart!

How Smart Speakers Can Detect a Heart Attack? As the research team reports, people who suffer from a heart attack often produce distinct sounds. According to the study, so-called agonistic breathing or wheezing is clearly distinguishable from normal breathing and can be detected by artificial intelligence. This is particularly important if sufferers experience a seizure during sleep or if they are alone in the apartment. A dispatch emergency call could, in these cases, double or even triple the chances of survival.

Rapid breathing increases the chances of survival

According to the research team, agonist breathing occurs in about 50% of people suffering from an acute heart attack. "This kind of breathing occurs when a patient experiences a really low oxygen level," explains Dr. Jacob Sunshine, one of the study's authors. Agonal breathing is characterized by a wheezing sound that is well suited as an audio biomarker for cardiac arrest. At the same time, the occurrence of pressure breathing is associated with a higher chance of survival, compared to those who respond in a cardiac arrest with apnea (respiratory arrest).

AI was developed with the help of emergency calls

The researchers used actual emergency call recordings to train AI. The system detected 97% of all snapshots at a distance of up to six meters. "Many people have smart speakers at home, and these devices have incredible features that we can use," says study co-author Professor Shyam Gollakota. The AI ​​can continuously and passively monitor the health of a person's heart without touching it. When the device detects an agonistic breath, the speaker informs the affected that an ambulance is being called. The user then has a small window of time in the unlikely event of a false alarm to cancel the emergency call.

Check for noise

Researchers attach particular importance to the fact that noise does not cause distortion in detection. "We've added a lot of irritating sounds to our tests, like the sounds of cats and dogs, car horns, air conditioners, and other things you normally hear in a house," adds the first dr. Justin Chan.

Rate of false positives can be reduced to 0%

In the first tests with audio recordings, the AI ​​falsely recognized a rapid breathing in 0.22 percent of the cases, although none existed. The team therefore introduced a safety detection that checks every ten seconds whether there is agonist breathing. This reduced the false positive rate to zero percent. According to the researchers, the AI ​​is not only suitable for smart speaker systems but also for smartphones and has already been tested on an iPhone 5s and a Samsung Galaxy S4. Currently, the concept is being tested for fitness for the general public on emergency calls in Greater Seattle. (Vb)


This article is based on the following sources:

Source link