The proof of concept tool using artificial intelligence was
developed by scientists at the University of Washington, in a study led by
first author Justin Chan, a doctoral student and co-corresponding author Shyam
Gollakota, an associate professor at Paul G. Allen School of Computer Science
& Engineering, University of Washington.
The findings of the study appear in
npj Digital
Medicine.
What is Agonal Breathing?
Agonal breathing occurs when a patient experiences really
low oxygen levels and sounds like a sort of a guttural gasping noise, which is
characteristic making it a good audio biomarker for someone having a cardiac
arrest.
Training AI
Tool to Identify Agonal
Breathing of Cardiac Arrest
- The
team collected sounds of agonal breathing from actual 911 calls to
Seattle's Emergency Medical Services
- Since patients
suffering from cardiac arrest are often unconscious,
bystanders recorded the agonal breathing sounds by placing their phones
near the patient's mouth while calling emergency services
- The team obtained
recordings from 162 calls between 2009 and 2017 and extracted 2.5 seconds
of audio at the beginning of each agonal breath and came up with 236 audio
clips
- They then
captured the audio clips on various smart devices including an Amazon
Alexa, a Samsung Galaxy S4 and iPhone 5s and employed machine learning
technology to increase the dataset to 7316 positive audio clips
- The team played
the audio clips from different locations in the bedroom to simulate how
the sound would be from various distances and also included interfering
sounds such as cars honking, the
sound of the air conditioner or a dog's bark that would
normally occur in a home
- This was done for
comparison so that the AI system could differentiate positive and negative
agonal breathing accurately, the team fed a negative dataset comprising 83
hours of audio data collected during sleep studies, and obtained a total
of 7,305 sound samples. These clips contained usual sounds people make
during sleep, such as snoring or obstructive sleep apnea
- Using the
positive and negative data sets, the research team employed machine learning
to produce a tool that could
identify agonal breathing 97% of the time when the smart speaker was
located up to 6 meters away from the person sleeping
- Additionally, the team tested the algorithm to make sure that the tool would
not identify normal heavy breathing or snoring mistakenly as agonal
breathing
"We don't want to alert either emergency
services or loved ones unnecessarily, so it's important that we reduce our
false positive rate," said Chan.
- During testing
the false positive rate of a breathing sound was 0.14% and false
positivity for separate audio clips recorded by normal volunteers was
0.22%
- When the team programmed the tool classify agonal breathing only
when it heard two distinct events at minimum 10 seconds interval, the
false positivity fell to zero.
The findings of the study suggest that
AI working through
smart speakers or smartphone apps could be potentially
lifesaving in cardiac arrest patients. The team feel this tool
could be used as a smartphone app or a skill for Alexa that runs in the
background while the person is asleep.
Future Plans
- Improving the
accuracy of the algorithm further and make it work across a larger
population by obtaining more 911 call data
- Commercialize
this tool through a University of Washington spinout, Sound Life Sciences,
Inc.
In summary, the findings of the study show how
AI tools
can be used in our daily lives to detect possibly life-threatening
symptoms and alert emergency services for prompt help.
References : - Contactless cardiac arrest detection using smart devices - (http://dx.doi.org/10.1038/s41746-019-0128-7)
Source: Medindia
Careers
