Image from CDC.gov
A few weeks after quarantine began in March, I woke up feeling under the weather. I had a headache, a stuffy nose, and a sore throat that was making me cough. Under normal circumstances, I'd recognize it as seasonal allergies rearing their ugly head, but considering the pandemic I immediately went to, "Oh crap, is this Covid?!"
Fortunately for me, a brief nap at lunchtime shook most of the symptoms and when I woke up at 100% the next day, I realize I had perhaps panicked for nothing. Imagine my relief (and, admittedly, disbelief!) to hear that researchers at the Massachusetts Institute of Technology (MIT) have developed a method for determining whether a person has COVID-19 just from the sound of their cough.
The tool uses neural networks that can detect the changes in a person's cough that are too subtle for the human ear to discern. The team at MIT developed an AI that can determine the minute differences by using tens of thousands of recorded samples of coughs and spoken words. Early tests have shown a 98.7% accuracy from people with confirmed cases, and 100% accuracy in the coughs of asymptomatic testers.
This technology could determine whether an individual is infected or not, even if they're not exhibiting any other symptoms. This is an impressive and important development, because it could tell a person who is otherwise feeling fine that they are an asymptomatic carrier of the virus.
The way it works is pretty neat. One of the neural network gauges sounds associated with vocal cord strength, while another detects cues related to a person's emotional state, like anger, sadness, anxiety, and so on, which produces a "flat effect." A third network then listens for subtle changes in lung and respiratory performance. All of the models combined were overlaid with an algorithm designed to detect muscular degradation.
The research is not, however, without its limits. Even with the level of accuracy demonstrated thus far, the AI is not a substitution for being tested for COVID-19. The technology also isn't built to diagnose people who are actively exhibiting symptoms of the virus. Despite this, the AI could become a valuable pre-screening tool for those who exhibit no symptoms but are concerned that they may be infected.
What do you think about this, is it something you would use if it were made available? Why or why not?
Learn more about the research on MIT's website.