In the world, up to 7 million people die from cardiac arrest each year. A new application of artificial intelligence could save lives. She works with smart speakers.
Nearly 500,000 Americans die each year from cardiac arrest. For German medical societies, name 65,000 cases in the same period. Every third of those affected are under the age of 65 and often, no risk factor was known before the event. Since many cardiac arrests occur outside hospitals, medical first responders lack resuscitation.
Researchers at the University of Washington have therefore developed a new program to monitor people without contact, for example during their sleep. You write, many cardiac arrests occur in the bedroom. Thanks to modern technology, help could be in sight: smart speakers such as Google Home and Alexa from Amazon, as well as newer smartphones, can analyze breathing sounds and, in the worst case, get even automatically help.
Digital assistants should give life to save
Background: In 50% of patients with cardiac arrest, there is talk of jerky breathing. "This type of breathing occurs when a patient has a very low oxygen content," says Jacob Sunshine. The Assistant Professor in Anesthesiology and Pain Medicine at the University of Washington School of Medicine added: "It's a guttural wheeze and the noise is so characteristic that it's a audio biomarker to determine if a person has cardiac arrest. "
But how should it go at home? "Many people have smart speakers and these devices have incredible capabilities that we can use," says Gollakota. He works at the Paul G. Allen School of Computer Science and Engineering at the University of Washington. "We are considering a contactless system in which the bedroom is continually and passively controlled to detect critical breathing noises and all nearby people are alerted by system messages." If they do not answer, a message at the center of emergency calls is possible.
Programming with calls from the emergency call center
To implement several steps were necessary. The researchers initially collected breath sounds on real emergency calls from the Seattle emergency services. Because most cardiac arrest patients are unconscious, their loved ones have their breath held by holding their phone against the patient's mouth. Between 2009 and 2017, the team collected exactly 162 phone calls involving patients in cardiac arrest.
The audio files were read and recorded with an Amazon Alexa, an iPhone 5s and a Samsung Galaxy S4. "We performed these examples at different intervals to simulate what it looks like when the patient is in different parts of the bedroom," says Justin Chan of the University of Washington. "We also added a lot of noise, from cats and dogs to car horns to air conditioners, to everything you normally hear in a house."
Minimize error rates – with artificial intelligence
In the evaluation, the researchers used artificial intelligence. First, the respiratory sounds of patients known to have suffered cardiac arrest were stored in a database. Noise has been added. Using algorithms, scientists used programs to search for specific patterns in the signal. This software was then tested with real data, the results being continuously optimized, keyword "Machine Learning". The result was a tool that correctly recognized fast breathing in 97% of cases. The distance between the device and the patient in the simulation was 6 meters maximum.
The researchers then tested their algorithm to make sure that other sleep sounds are not confused with jerky breathing. After all, false positive signals would quickly lead to the end of all developments. Depending on the setting, the false positive rate was between 0.14% and 0.22%. As soon as your software is considered an instant event, if there were two identical readings within ten seconds, the rate would go to 0%.
From the laboratory to the application
After the feasibility study, it should evolve rapidly towards the application. The team imagines that the algorithm could work as an application or a "skill" Alexa. The application must be activated before sleeping. This circumstance is a possible obstacle: after all, digital assistants are criticized because they too often have their "ear". Since many at-risk patients are not known as such, they should hardly benefit from the new technology. It was known, therefore, that patients were at risk of cardiac arrest.
More about medicine: