In an effort to tackle in-home cardiac arrest, University of Washington researchers have devised a novel contactless system that uses smartphones or voice-based personal assistants to identify telltale breathing patterns that accompany an attack.
The proof-of-concept strategy, described in an NPJ Digital Medicine paper published this morning, involved a supervised machine learning model called a support-vector machine that was trained for use in the bedroom, a controlled environment in which the majority of in-home cardiac arrests occur.
“Sometimes reported as ‘gasping’ breaths, agonal respirations may hold potential as an audible diagnostic biomarker, particularly in unwitnessed cardiac arrests that occur in a private residence, the location of [two-thirds] of all [out-of-hospital cardiac arrests],” the researchers wrote. “The widespread adoption of smartphones and smart speakers (projected to be in 75% of US households by 2020) presents a unique opportunity to identify this audible biomarker and connect unwitnessed cardiac arrest victims to emergency medical services (EMS) or others who can administer cardiopulmonary resuscitation.”
TOPLINE RESULTS
Cross-validation analysis of the trained classifier yielded an overall sensitivity and specificity of 97.24% and 99.51%. Running the classifying over audio streams that excluded positive cases used to train the tool yielded a false positive rates of 0.14% and, when a frequency filter was applied to the algorithm, 0.00085%.
However, to gauge the tool’s real-world performance, the researchers conducted another test involving 35 individuals who recorded themselves sleeping for a total of 167 hours with their smartphones. After retraining the classifier with a sample of data from each participant, the tool then demonstrated a sensitivity and specificity of 97.17% and 99.38%, with unfiltered and filtered false positive rates of 0.22% and 0.001%.
HOW IT WAS DONE
The team’s system captures audio samples using the microphone of a smartphone or speaker, and generates a real-time probability of agonal breathing for each 2.5-second segment.
To train their support-vector machine model, the researchers sourced audio from recorded 911 emergency calls, among which 162 contained a clear recording of agonal breathing. To bolster the relatively small dataset, the researchers played these recordings over the air at various distances, and with interfering indoor and outdoor sounds. These recordings were captured by an Amazon Alexa, iPhone 5s and Samsung Galaxy S4, yielding a total of 7,316 positive 2.5-second samples for training.
Negative samples were collected from 83 hours of sleep study recordings from 12 different patients, many of which included other instances of irregular breathing or other bedroom noises that could act as interference.
WHAT’S THE HISTORY
A contactless detection system may be novel within the realm of in-home cardiovascular disease, but the approach is already making waves in sleep monitoring and fall detection. Examples of the former include apps that monitor sleeping sounds through a smartphone’s microphone, while the latter is a staple feature of Kardian's and Vayyar’s offerings.
IN CONCLUSION
The team noted that their take on this system ran on the devices locally and without saving any data, due in part to privacy concerns and the need for continuous monitoring. And while this proof-of-concept’s training and implementation were limited, the authors were still enthused by the opportunity a contactless, consumer device-friendly detection tool might have for preventing the nearly 300,000 out-of-hospital cardiac arrest deaths reported in North America each year.
“Technology is rapidly evolving and in turn providing opportunities to improve human health. The increasing adoption of commodity smart speakers in private residences and hospital environments may provide a wide-reaching means to realize the potential of a contactless cardiac arrest detection system,” the researchers concluded.
