Through the sound of the cough recorded by a mobile phone, the algorithm is able to detect asymptomatic cases of COVID-19
Since April, the Massachusetts Institute of Technology (MIT) has developed an artificial intelligence-based algorithm for COVID-19 detection through coughing. After months of training the algorithm through 70 thousand recordings of forced cough that sent volunteers through mobile devices or web platforms the process was completed. For data collection, MIT's partnerships with universities and study centers around the world, such as Tose con causa (https://www.toseconcausa.udg.mx) campaign, launched by the University of Guadalajara in June, were important.
MIT researchers found that people with an asymptomatic SARS-CoV-2 infection can cause changes in people's coughs, so a person's cough with asymptomatic infection is different from that of an infection-free person. However, the differences between the two coughs are not noticeable to the human ear, in the face of this, Artificial Intelligence was key, to develop a reliable algorithm that could verify the assertion.
The algorithm was fed tens of thousands of cough samples as well as spoken words. The algorithm accurately identified 98.5% cough of people with confirmed COVID-19 case, and 100% cough of asymptomatic people.
MIT's plan with this innovation is to incorporate this AI model into an easy-to-use application and seek FDA (Food and Drug Administration) approval to make it available to the population for free and easily identify asymptomatic people, and reduce contagion.
“The effective implementation of this group diagnostic tool could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant,” said Brian Subirana, a Catalan research scientist who is a member of MIT's Auto-ID Laboratory and developer of this project.
Prior to the pandemic, researchers had worked on training algorithms that could diagnose pneumonia or asthma through mobile cough recordings. On the same line MIT had developed AI models for cough analysis for Signs of Alzheimer's. “The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs. 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,” Subirana explained.
Thanks to previous studies, they began testing this type of biomarkers in COVID-19. It was in April that they began collecting voice recordings from both confirmed COVID-19 patients and healthy people.
The 70.000 recordings contain various types of cough, which equates to about 200 thousand cough samples. Nearly 2.500 recordings were for people who confirmed coVID-19, including asymptomatic ones.
Researchers Jordi Laguarta and Ferran Hueto accompanying Subirana agreed that there are similarities between Alzheimer's discrimination and COVID-19. Since they didn't have to make many adjustments to the AI model that was originally built for Alzheimer's. This model detects patterns in four biomarkers: vocal cord strength, feeling, lung and respiratory performance and muscle degradation.
Advances in his research were published in IEEE Journal of Engineering in Medicine and Biology. Finally, the authors mention that “pandemics could be a thing of the past if pre-screening tools are always on in the background and constantly improved.”
