At least one in five people who get Covid-19 will not show any symptoms and may unwittingly pass the virus on to others. Those who do not feel sick and are unaware of their contact with an infected person cannot know that they should get tested.
However, researchers at the Massachusetts Institute of Technology (MIT) may have found a way to identify these symptom-free coronavirus carriers without a test.
In a study published in September, researchers tested an artificial intelligence that can distinguish between the coughs of people with Covid-19 and healthy people. She was even able to tell whether people were healthy or asymptomatic carriers by feigning or forced coughing – based on sound variations too subtle for the human ear to perceive.
AI has not misdiagnosed a single cough from an asymptomatic carrier
Using the MIT model, 98.5 percent of people in the study infected with the coronavirus, including asymptomatic carriers, could be detected because of their cough. In fact, the AI has not misdiagnosed a single cough from an asymptomatic carrier.
“We think this shows that the way you make sound changes when you have Covid, even when you’re symptom-free,” said Brian Subirana, scientist and one of the study authors, in an MIT press release.
Subirana and his colleagues suggested that their AI could be built into speakers and cell phones. “Pandemics could be a thing of the past if the pre-screening tools were always running in the background and constantly being improved,” said Subirana and his colleagues.
The way we cough is emotional
Before the pandemic, the researchers had trained AI models to recognize other diseases such as Alzheimer’s and pneumonia based on people’s coughs.
This is possible because the way we speak and cough reflects the strength of our vocal cords and the organs around them, Subirana said. Therefore, for example, Alzheimer’s patients tend to have weaker vocal cords due to neuromuscular impairments.
“Artificial intelligence can detect things that we can easily deduce from language simply by coughing, including things like gender, mother tongue or even the emotional state of the person. In fact, there is a feeling in the way you cough, ”Subirana said.
So Subirana’s team applied these earlier models to Covid-19 patients. They collected 70,000 audio samples from coughing people, both healthy and infected, and asked the latter group to report any Covid-19 symptoms or absence thereof.
More than 2,600 of the recordings were submitted by people who had tested positive for Covid-19. Then the researchers had the artificial intelligence listen to about 4,250 of these recordings, including those of infected people.
The artificial intelligence was able to create an acoustic picture of what a sick and healthy cough sounds like. She identified specific patterns in vocal cord strength, lung performance, emotions, and muscle breakdown in coughers that were unique to Covid-19 patients.
Once the AI was done, Subirana’s team listened to more than 1,000 coughing signals. The AI identified 100 percent of coughs that are attributable to asymptomatic Covid-19 carriers.
A tool that should be used “before going to a classroom, factory or restaurant”
The Subirana team is working on integrating the AI model into a free app and is also hoping to get approval from the U.S. Food and Drug Administration (FDA) to use AI as a Covid-19 early detection tool.
So in theory, a person could cough into their phone and then immediately find out if they might be a symptom-free carrier of the coronavirus. However, the person would still need to take a test to confirm the diagnosis of the AI. This type of early detection tool “could reduce the spread of the pandemic if everyone uses it before going to a classroom, factory or restaurant,” Subirana said.
However, the AI cannot determine whether the cough is due to another disease such as Covid-19, particularly the flu or a cold. It was developed solely to detect coughs from Covid-19 carriers.
The study results are encouraging, according to Anthony Lubinsky, director of respiratory care at New York University’s Langone Tisch Hospital. But he told the US news platform Live Science that “further research is needed to determine whether the AI works well enough in practice to recommend its use as a screening tool.”
The study’s authors report that they have already worked with hospitals in the United States, Mexico, and Italy to collect more cough records – and so keep improving and testing their model.
This text has been translated from English. You can find the original here.