An AI model can detect asymptomatic COVID-19 infection by listening to a patient’s cough recorded on a mobile phone.
Researchers at MIT (Massachusetts Institute of Technology) found that the distinction between a COVID and non-COVID cough was not clear to the human ear but could be picked up using machine learning.
They built up a database of 5,320 recordings between April and May 2020.
The researchers trained the model on 4256 cough samples. They then tested it on the remaining 1064 subjects.
The model accurately identified 98.5% of coughs from people confirmed to have COVID-19.
This included 100% of coughs from people who were asymptomatic but had tested positive for the virus.
The team, who are currently working on developing an app, say their technique could produce a free, non-invasive, real-time, large-scale COVID-19 testing tool.
One of the authors, Brian Subirana, a research scientist at the MIT auto-ID laboratory said: “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.”
A Pdf of COVID-19 Artificial Intelligence Diagnosis using only Cough Recordings is available here