The Massachusetts Institute of Technology (MIT) has developed an algorithm to identify people who have been infected with COVID-19. The algorithm listens to your footsteps and understands whether you are infected or not.
The algorithm was trained using “tens of thousands” of recordings – both coughs and spoken words – and was able to identify 98.5% of those who showed symptoms and were confirmed cases of COVID-19.
Additionally, the algorithm identified 100% of COVID-19 carriers who were confirmed to have the virus but did not show symptoms.
The recordings used to train the artificial intelligence (AI) model were submitted by volunteers online and included forced coughing from healthy volunteers as well as from COVID-19 patients. Over 70,000 samples have been collected so far, and approximately 2,500 have been submitted from people confirmed to have COVID-19.
“People who are asymptomatic carriers of COVID-19 have a different cough than healthy individuals,” the team commented. “These differences are not detectable by the human ear. But it turns out that they can be deciphered by artificial intelligence.”
The MIT team, consisting of Brian Subirana, Jordi Laguarta, Ferran Hueto from MIT's Auto-ID lab, says they are now working on a user-friendly mobile app that will incorporate the algorithm.
However, such an app requires approval and should not be considered an official diagnostic tool. Instead, the app could potentially act as a “non-invasive screening” for users before confirming any suspicions with a diagnostic test. It should also be kept in mind that an asymptomatic cough can be associated with the flu, colds, or other conditions.
Any tool that could potentially address the issue of asymptomatic transmission – in which those without symptoms could inadvertently spread the virus – could be valuable in the fight against COVID-19.
“Effective implementation of this group diagnostic tool could reduce the spread of the pandemic if people used it before going to school, the factory or the restaurant,” says Subirana.
MIT is now working with some hospitals that will provide additional step recordings to further refine the model.
The AI model builds on previous studies that explored how coughing and changes in speech patterns could indicate other diseases, including Alzheimer's.
