This AI Identifies Asymptomatic COVID Carriers by Their Cough
A neural network created by MIT to look for signs of Alzheimer’s has been applied to COVID-19 and accurately identified 98.5 percent of coughs from people who were confirmed to have novel coronavirus.
Not everyone who contracts COVID-19 will develop noticeable symptoms: They don’t run a fever, feel short of breath, or lose their sense of smell or taste. So how do you identify asymptomatic carriers? It’s in their cough, according to the Massachusetts Institute of Technology.
In a paper published recently by the IEEE Journal of Engineering in Medicine and Biology, a team of MIT researchers suggests that differences indecipherable to the human ear may, in fact, be recognized by artificial intelligence.
An AI model, trained on tens of thousands of sample coughs and spoken words, was able to accurately identify 98.5 percent of coughs from people who were confirmed to have novel coronavirus—including 100 percent of those who were not displaying symptoms. Asymptomatic carriers are more likely to continue socializing, shopping, and potentially spreading the disease than those experiencing chills, fatigue, body aches, nausea, and other demonstrable markers of COVID.
MIT researchers were already training algorithms pre-pandemic to analyze forced-cough recordings for signs of Alzheimer’s. The combined neural network also looks for different degrees of vocal cord strength and emotional states in speech—all of which, when considered together, serve as signatures for diagnosing senile dementia.
When coronavirus began spreading, the team wondered whether their existing AI could lend a hand. “There’s in fact sentiment embedded in how you cough,” paper co-author Brian Subirana, a research scientist in MIT’s Auto-ID Laboratory, explained. “So we thought, why don’t we try these Alzheimer’s biomarkers [to see if they’re relevant] for COVID.”
Spoiler alert: they are. Research revealed a “striking similarity” between the two diseases, and the AI required little tweaking to pick up patterns specific to novel coronavirus. “We think this shows that the way you produce sound changes when you have COVID, even if you’re asymptomatic,” Subirana said.
The MIT team is working to incorporate their model into a mobile app that could serve as a free, convenient, non-invasive pre-screening tool for COVID-19. A user, for example, could log in daily, cough into their phone, and instantly learn whether they might be infected and should take a formal test. “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,” according to paper co-author Brian Subirana, a research scientist in MIT’s Auto-ID Laboratory.
A commercial version of the AI, however, is not meant to diagnose symptomatic people; it is simply another tool to distinguish asymptomatic coughs from healthy ones.