Researchers Develop AI That Can Check Cough Sounds For Early Detection Of COVID-19

Illustration of artificial intelligence for health (Photo: Antara)

JAKARTA - Researchers from the Massachusetts Institute of Technology (MIT) developed artificial intelligence (AI) technology that can check the sound of a cough which can be an early warning that someone is most likely in the early stages of being infected with COVID-19.

Quoted Antara from the Tech Crunch page, Sunday, November 1, previously AI has been made to detect conditions such as pneumonia, asthma, and even neuromuscular disease. Before the pandemic, researcher Brian Subirana had shown that coughing could even help predict Alzheimer's.

Recently, Subirana thought that AI could tell so many things, including COVID-19. He and his team created a website where people could contribute their coughing voice to research data. Thousands of samples are used to train this AI.

The AI detected subtle patterns in vocal strength, lung and respiratory performance, and muscle degradation, thus identifying 100 percent of coughs by asymptomatic COVID-19 carriers and 98.5 percent with symptomatic, with specificities of 83 percent and 94 percent, respectively, which means the results are pretty accurate.

"We think this shows that the way you produce sound changes when you get COVID, even if you don't have symptoms," Subirana said of the findings.

However, he cautioned that while these systems are good at detecting unhealthy coughs, they should not be used as a diagnostic tool for people with symptoms but are unsure of the underlying cause.

"The tool detects a feature that allows it to distinguish between subjects who have COVID and those who don't," Subirana told Tech Crunch, explaining further.

"Previous research has shown that you can have other conditions too. One can design a system that will differentiate between many conditions, but our focus is on choosing COVID," he added.

For those who pay attention to statistics, 100 percent is not a number often seen in AI models. These findings need to be proven in other data sets and verified by other researchers.

The Subirana team is working with several hospitals to build a more diverse data set. The research team is also working with private companies to develop applications to distribute the tool for wider use, if it gets approval from the United States Food and Drug Administration, FDA.