[ad_1]
MIT researchers have trained artificial intelligence to detect people infected with Covid-19 without symptoms anyway: through an analysis of their cough.
SarsCov2 is a particularly insidious virus: many infected people, especially young people, do not show any typical symptoms. But they are still contagious. And there is another problem: even those who really get sick are contagious even before the first symptoms.
Many infected people shed the virus a day or two before the first signs of the disease. The intentional cough of a healthy person and the intentional cough of a coronally infected person without symptoms can hardly distinguish even trained ears like doctors.
People with Corona infection cough differently
MIT AI can do just that, at least that’s what the researchers say in their study in the Journal of Engineering in Medicine and Biology. Scientists have collected more than 70,000 cough records from volunteers – a huge database.
Telltale cough, even without symptoms
2,500 cough samples were from people known to be infected with SarsCov2. Some had symptoms and some didn’t. AI software learned from audio samples to correctly distinguish coughs – for those infected with no symptoms, this succeeded at an astonishing 100 percent. For Covid patients with symptoms, the success rate was slightly lower, at 98.5%.
The algorithm evaluates four different factors: vocal cord strength, lung function, emotional state, and muscle breakdown. Apparently, SarsCov2 leads to measurable changes here, even if the infected person doesn’t even notice.
Artificial intelligence feeds on coughing
Researchers are now collecting more cough records in clinics around the world in order to train AI even better. The goal is to create an app that allows quick diagnosis via smartphone or smart speaker. The MIT inventors themselves warn against exaggerated expectations: their software is primarily intended to detect infected people WITHOUT symptoms. AI was not developed to distinguish acute Covid cough from flu cough or bronchitis.
.
[ad_2]
Source link