Researchers at MIT have discovered an unlikely way of discerning those with COVID-19 from healthy individuals — simply by the way they cough. The coughs of COVID-19 patients possess extremely subtle differences in their sound signatures, too faint for the human ear to detect. Thanks to machine learning, computers, however, can be trained to tell these sounds apart.
A newly-developed diagnostic platform powered by artificial intelligence described in the IEEE Journal of Engineering in Medicine and Biology has demonstrated its ability to positively pick out COVID-19 patients based solely on the sound of forced-cough recordings. This could mean that future COVID-19 testing could be as simple as picking up the telephone.
Such AI-driven diagnostic platforms bring with them immense promise. These systems analyze human-to-machine interactions, understanding requests, connecting data points, and automatically drawing connections much more efficiently and accurately than human beings would be able to.
To build the diagnostic algorithm, known as ResNet50, the team at MIT input tens of thousands of audio recordings of coughs and spoken words from both healthy and COVID-19 infected individuals. When they put the system to the test by playing it new recordings, the results were remarkable: it identified 98.5 percent of COVID-19 patients correctly based on their coughs. Even more phenomenal was the system’s ability to pinpoint asymptomatic COVID-19 patients (those who tested positive but were not showing the typical symptoms of cough and fever) 100 percent of the time.
“The effective implementation of this group diagnostic tool could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant,” said study co-author Brian Subirana from MIT’s Auto-ID Laboratory.
The team is actively working on the next steps for the technology — a smartphone app. After further validation and approval by regulatory bodies, this app could transform the way COVID-19 cases are identified and tracked. Users would have access to a convenient and completely noninvasive COVID-19 prescreening tool, all in the palms of their hands. The app would provide instant guidance on whether a person might be infected and they could follow up with clinical intervention much faster, avoiding the spread of the coronavirus and the escalation of symptoms.
Source: MIT News, IEEE Journal of Engineering in Medicine and Biology.