A team of University of Washington researchers has been training artificial intelligence to identify instances of cardiac arrest and respond accordingly.
As part of the project, the team has trained an AI system to study 911 call samples to recognize the distinct gasps for air that are let out by someone going into cardiac arrest. Initially, the system will ask nearby people for help to provide CPR, although it will call 911 shortly if no response is received.
The researchers say they used 162 calls collected between 2009 and 2017 and extracted 2.5 seconds of audio at the start of each agonal breath to come up with a total of 236 clips. For greater accuracy, the team captured recordings through various devices, including Amazon Alexa, an iPhone 5S and a Samsung Galaxy S4, and bolstered their abilities through machine learning.
Ultimately, the AI only misidentified breathing 0.22 percent of the time or less when it detected a single event and had perfect detection when listening for events at least 10 seconds apart, according to the team.
Going forward, the team eventually intends to commercialize this tech through a spinout company called Sound Life Sciences. The researchers envision their algorithm being used in an app or virtual assistant skill that runs passively on a smart speaker or smartphone while people are sleeping.