- A recent study found that an artificial-intelligence model could tell the difference between a regular cough and a COVID-19 cough — even forced coughs from asymptomatic carriers.
- The AI identified 98.5% of the coughs from people with coronavirus infections, asymptomatic or otherwise.
- The researchers hope to incorporate the model into an app, as well as smart speakers and cell phones, as a real-time screening tool.
At least one out of every five people who get the coronavirus doesn’t show symptoms and can unknowingly spread the virus to others. Those who don’t feel sick and aren’t notified of exposure can’t know that they should get tested.
But researchers at the Massachusetts Institute of Technology may have found a way to identify these silent coronavirus carriers without a test.
study published in September describes an artificial-intelligence model that can distinguish between the coughs of people with the coronavirus and those who are healthy. It can even tell from voluntary, forced coughs whether people were healthy or were asymptomatic carriers, based on sound variations too subtle for the human ear to discern.
The MIT model accurately detected 98.5% of people in the study who had coronavirus based on their coughs, including asymptomatic carriers. In fact, the program didn’t misdiagnose a single cough from an asymptomatic carrier.
“We think this shows that the way you produce sound changes when you have COVID, even if you’re asymptomatic,” Brian Subirana, a research scientist and coauthor of the study, said in an MIT news release.
Subirana and his colleagues suggested that their AI could be incorporated into smart speakers and cell phones.
“Pandemics could be a thing of the past if pre-screening tools are always on in the background and constantly improved,” they wrote.
‘Sentiment embedded in how you cough’
Before the pandemic, the researchers had been training AI models to detect other diseases like Alzheimer’s and pneumonia based on people’s coughs. This is possible because the way we talk and cough can reflect the strength of our vocal cords and the organs surrounding them, according to Subirana. Patients with Alzheimer’s, for example, tend to have weaker vocal cords due to neuromuscular degradation.
“Things we easily derive from fluent speech, AI can pick up simply from coughs, including things like the person’s gender, mother tongue, or even emotional state. There’s in fact sentiment embedded in how you cough,” he said.
So Subirana’s team applied those earlier models to COVID-19 patients.
They collected 70,000 audio samples of coughs from both healthy and infected people, and asked the latter group to report any coronavirus symptoms or lack thereof. More than 2,600 of the recordings were submitted by people who’d tested positive for COVID-19.
Then the researchers had their artificial-intelligence model listen to about 4,250 recordings, including the audio from infected people. The AI was able to build a picture of what sick and healthy coughs sound like. It identified specific patterns in coughers’ vocal cord strength, lung performance, emotions, and muscular degradation that were unique to COVID-19 patients.
Once the model was ready, Subirana’s team then had it listen to more than 1,000 coughs. It identified 100% of the coughs belonging to asymptomatic coronavirus carriers.
A tool to be used ‘before going to a classroom, a factory, or a restaurant’
Subirana’s team is working on incorporating the AI model into a free app, and also hoping to get authorization from the Food and Drug Administration for its use as a COVID-19 pre-screening tool.
Theoretically, a person could cough into their phone then immediately find out whether they may be an asymptomatic coronavirus carrier. They’d still need a test to confirm a diagnosis, though.
This kind of a pre-screening tool “could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant,” Subirana said.
However, the AI can’t determine whether your cough is due to a different illness in particular, like the flu or a cold. It’s only calibrated to detect COVID-19. It also can’t say whether the fever or sore throat you have is due to COVID-19.
According to Anthony Lubinsky, director of respiratory care at New York University Langone Tisch Hospital, the study results are encouraging.
told Live Science that “whether or not this performs well enough in a real-world setting to recommend its use as a screening tool would need further study.”
The study authors said they’ve already partnered with hospitals in the US, Mexico, and Italy to collect more cough recordings to further improve and test their model.