[ad_1]
An application recognizes Covid-19 when coughing: thanks to artificial intelligence, massive free tests are possible
ETH Lausanne is developing self-learning algorithms to detect Covid-19 by coughing on a smartphone.
Click “Burn” and then cough, please. You are asked to do so on the EPFL’s Coughvid website (EPFL), where the researchers collect cough noises. Because the type of cough can reveal if someone has Covid-19.
The Coughvid project, launched in April by EPFL’s Integrated Systems Laboratory, uses artificial intelligence (AI) to try to slow the spread of Sars-CoV-2. Using signal processing and machine learning, an Android app and website will be developed “to automatically and conveniently monitor Covid-19 from home,” as the researchers write.
Covid-19 cough is different from normal cough
Recognizing Covid-19 by coughing sounds quite adventurous. But the Coughvid researchers write that, according to the World Health Organization, 68 percent of all Covid-19 patients have a dry cough. So there is no wet cough as with a cold or due to an allergy. And the sound of a dry cough is very different from a wet cough. Researchers at EPFL have confirmed this with robust analysis of other respiratory diseases such as asthma, pneumonia and whooping cough.
Researchers at the Massachusetts Institute of Technology in Boston (MIT) are also working on an application for the detection of Covid-19. They found that there are even differences between the common cough and the cough of people infected with Covid-19 if the cough was forced. Humans are unable to hear this difference, but artificial intelligence can.
Thanks to machine learning, the recognition rate improves
Both research groups are based on machine learning with a large data set of cough tones from healthy people and people infected with Covid-19. Therefore, the Lausanne researchers ask everyone to send a few seconds of cough tones through their smartphone. This is how the algorithm is trained to differentiate between Covid-19 cough tones. In its tests, MIT reports that it correctly identified 98.5 percent of those infected with corona. For subjects who had no symptoms, the detection rate was even 100 percent, as the MIT researchers write in the “IEEE Open Journal of Engineering in Medicine and Biology.” Therefore, the method is also suitable for infected people without symptoms that can already infect other people.
MIT is still ahead of the game
EPFL researchers have not yet reached this recognition rate. In the “Netzwoche,” Lausanne research director Tomas Teijero says that his model is currently able to recognize 40 percent of those infected from the noise of coughing. And only in three percent of cases is a person wrongly recognized as positive.
That’s not enough. With more cough tones sent, the adaptive algorithm will increase the recognition rate. The Lausanne researchers have teamed up with Cambridge scientists to exchange data sets. If the application runs reliably, you can easily and free test your smartphone in the future to see if it is infected or not. For example at the entrance of the school, the stadium or the office.
Participating is easy
It’s easy to participate, by the way: mark your health status on the Coughvid app or website, enter your age and gender, then cough and send the recording to the EPFL server.
