Artificial intelligence taught to recognize Covid-19 by the sound of a cough
The model accurately identified 98.5% of people with confirmed Covid-19 by coughing.
People who are infected with COVID-19 but do not have symptoms show no discernible physical signs of illness. Thus, they are less likely to be tested for the virus and may unknowingly spread the infection to others.
However, even with an asymptomatic course of the disease, Covid-19 can be detected. Researchers at the Massachusetts Institute of Technology have found that asymptomatic patients can differ from healthy people by their cough. The human ear does not pick up these differences, but artificial intelligence very well distinguishes the coughing of those infected from coughing who are not at all sick.
The model was trained on tens of thousands of cough samples that participants sent to researchers. Subsequently, the model successfully identified the cough of confirmed coronavirus patients in 98.5% and asymptomatic patients in 100% of cases.
To date, researchers have collected more than 70,000 records, each of which contains several examples of coughs, about 200,000 samples in total. About 2,500 entries were submitted by people who had confirmed the presence of coronavirus, including those in an asymptomatic form.
Scientists have created a neural network that was trained using recordings made during phone calls or calls in various instant messengers. The models “fed” tens of thousands of cough audio, supplementing them with specially made recordings of a forced cough.
The study leveraged previous research that used forced cough recordings to predict the development of Alzheimer’s disease. This pathology is associated not only with memory impairment, but also with neuromuscular degradation, such as a weakening of the vocal cords.
After training, the neural network accurately identified 98.5% of people who had Covid-19 by coughing. Including the model identified 100% of asymptomatic patients whose test result for coronavirus was positive. Development article published в издании IEEE Journal of Engineering in Medicine and Biology.
Now a team of scientists is working to ensure that the program can be used through a mobile application. If the FDA (agency of the US Department of Health and Human Services) approves the development on a large scale, then the application can become a free and convenient tool for detecting disease. The user can access the application whenever they want, cough into their phone and instantly receive information about their condition.
The inventors emphasize that their development cannot be the only tool for diagnosing coronavirus. If the AI detects an infection, then it is necessary to pass the traditional test for Covid-19.