A new algorithm can accurately give a coronavirus diagnosis solely through the sound of a cough.
Developed by the Massachusetts Institute of Technology’s Auto-ID Laboratory, the artificially intelligent technology was recently tested on 2,500 people. It correctly identified 97.1% of those with confirmed cases of COVID-19 and 100% of those with asymptomatic cases.
Researchers behind the algorithm say it can spot differences between healthy coughs and coronavirus coughs better than humans can. They don’t see it as a diagnostic tool, but rather an early warning system that can be an affordable and non-invasive option to prescreen people for COVID-19 and tackle the problem of asymptomatic transmission.
“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,” MIT scientist Brian Subirana, who helped lead the research, said.
Subirana and his team hope to soon gain regulatory approval to embed the technology into a smartphone app for widespread usage. The algorithm utilizes ResNet50, a neural network initially developed to detect Alzheimer’s through human speech. The network gained precise accuracy through hours of training analyzing words spoken by humans in different physiological and emotional states. It’s gotten so good it can detect personality traits as well as gender, mother tongue, and mood solely through examining a person’s voice.
By spotting differences in lung and respiratory performance, the technology may soon become the most efficient resource available to quickly get a coronavirus diagnosis.
“The way you produce sound changes when you have COVID, even if you’re asymptomatic,” Subirana said. “The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs. This means that when you talk, part of your talking is like coughing, and vice versa.”
The technology’s nearly perfect success rate is motivating researchers to test the tool on a more diverse set of data to determine other contributing factors before looking at smartphone implementation.
Several similar projects are underway across the Atlantic. Cambridge University and U.K .health start-up Novoic have all been developing cough detection algorithms of their own.
“It’s an example of AI being helpful,” artificial-intelligence expert Calum Chace said. “For once, I don’t see a lot of downside in this.”