This AI can identify the coughs of asymptomatic people with Covid-19

Sometimes, artificial intelligence knows more about us than we know about ourselves. And when it comes to diagnosing Covid-19, that could be a pandemic-fighting boon.

Researchers from MIT have developed an algorithm they say can differentiate the forced coughs of asymptomatic people who have Covid from those of healthy people. They are currently working on a free app that would enable anyone to cough into their smartphone, and essentially get a pre-screening of whether they might have Covid, even if they don't have symptoms.

The researchers published their paper in the IEEE Journal of Engineering in Medicine and Biology, and the MIT News spoke with the team about the research to learn more.

Prior to the pandemic, the MIT team had been working on AI models that could diagnose diseases, notably including Alzheimer's, based on vocal recordings. Training the AI on 200,000 recordings of speech and coughs, they found that AI could detect biomarkers like vocal cord strength, lung capacity, and neuromuscular degeneration that show up in these recordings, and differ from the coughs and words of healthy people.

Since Covid can have some of the same symptoms (notably "temporary neuromuscular impairment"), the researchers decided to apply their Alzheimer's framework to detecting Covid-19. The idea is that the AI could pick up these differences in the coughs caused by neuromuscular impairment and other symptoms between healthy people and asymptomatic people with Covid-19, differences that those asymptomatic people would not be able to notice themselves.

According to their research, the theory worked. Since they had already trained the initial model on tens of thousands of recordings, they were able to create a Covid-specific assessment from 4,000 cough samples, half from healthy people, half from asymptomatic Covid-19 patients.

They then tested the AI on 1,000 further samples (again, half healthy, half not), and it correctly identified 98.5 percent of coughs from people with Covid-19. It accurately tagged all of the forced coughs from asymptomatic Covid-19 patients. Pretty impressive!

The researchers say more refinement of the AI model with more training samples needs to be done. However, they're hopeful that this could be a useful tool to fight the pandemic.

Research shows that asymptomatic Covid patients unknowingly contribute to the spread of the disease. A cough pre-screening tool that allows people to easily get a sense of whether or not they have Covid-19 before they unwittingly go out into the world and potentially infect others could combat that.

Imagine, before leaving the house, asking Alexa to check whether you should probably go get a Covid test by just letting Alexa listen to your coughs. That scenario might sound a little freaky, but if it works, we'll take it.