Coronavirus: Researchers create AI model that can detect infections through a cellphone-recorded cough

(Photo: Pixabay)
(Photo: Pixabay)
Pixabay
  • A team of researchers have developed a model using AI that can detect SARS-CoV-2 infections via a simple forced cough
  • If the model receives regulatory approval, it can help tell people, especially those who are asymptomatic, if they should isolate
  • However, the model cannot be used to diagnose Covid-19 – so far, only an official test can achieve this

A model developed by researchers from the Massachusetts Institute of Technology (MIT), using an artificial intelligence (AI) algorithm, can correctly detect people infected with Covid-19, using only the sound of a forced cough.

The tests achieved a 98.5% success rate among people who had officially tested positive for coronavirus, and 100% in asymptomatic cases.

Interestingly, the results also show that it can spot differences in coughing that can't be heard with the human ear.

Part of talking is like coughing

If the researchers receive regulatory approval to develop it into a user-friendly app, the non-invasive model could become a highly useful screening tool globally, especially for people who may not suspect they are infected.

"The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs," researcher Brian Subirana, from MIT said in the university's news release.

"This means that when you talk, part of your talking is like coughing, and vice versa.

"It also means that things we easily derive from fluent speech, AI can pick up simply from coughs, including things like the person's gender, mother tongue, or even emotional state. There's in fact sentiment embedded in how you cough."

Their findings were published in the IEEE Open Journal of Engineering in Medicine and Biology.

AI initially designed for Alzheimer's disease

The researchers initially created the model for the study of coughs and Alzheimer's disease, but later shifted their focus to the new coronavirus. However, they stated that their focus will shift back to their initial intention once the Covid-19 pandemic is behind us.

For this study, their work involved a neural network, known as ResNet50, that was trained on tens of thousands of human coughs, as well as spoken words, to spot changes in lung and respiratory performance.

After combining the three models, the team filtered out stronger coughs from weaker ones using a layer of noise.

They wrote that around 2 500 captured cough recordings of people confirmed to have the virus were used in the study.

"The effective implementation of this... tool could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant," said Subirana.

Not designed to diagnose Covid-19

However, despite the impressive findings, the team still has a long way to go in finding a way for the model to detect the differences between healthy coughs and unhealthy coughs in asymptomatic people.

More importantly, they cautioned that the model can only be used as an early warning system to inform people whether they should isolate based on the result, and that it cannot actually diagnose Covid-19.

“Pandemics could be a thing of the past if pre-screening tools are always on in the background and constantly improved,” the authors wrote.

READ | Seven different 'forms of disease' identified in mild Covid-19 cases - study

READ | Some Covid-19 patients may only show gastrointestinal symptoms, new review suggests

READ | Ethics and Covid-19: Who should get the ventilator when resources are scarce?

Image: Pixabay

We live in a world where facts and fiction get blurred
In times of uncertainty you need journalism you can trust. For only R75 per month, you have access to a world of in-depth analyses, investigative journalism, top opinions and a range of features. Journalism strengthens democracy. Invest in the future today.
Subscribe to News24
Voting Booth
Have you entered our Health of the Nation survey?
Please select an option Oops! Something went wrong, please try again later.
Results
Yes
34% - 9269 votes
No
66% - 17863 votes
Vote