The FluSense AI turns the sound of people coughing into useful data

Your coughing could help epidemiologists better understand the distribution pattern and transmission potential of viral respiratory outbreaks in crowds. University of Massachusetts Amherst researchers have created an artificial intelligence device called FluSense which has a rather simple but increasingly important job: analyze audio of people coughing in public places and turn it into comprehensible, actionable data.

How FluSense works — With the help of a relatively cheap microphone array and thermal imaging data set in a public area — researchers focused on four waiting rooms at the university's health services clinic — FluSense's creators mined and studied over 350,000 thermal images and 21 million non-speech audio clips between December 2018 to July 2019.

Using the datasets, FluSense was able to correctly predict illness rates with impressive accuracy. Researchers noted that FluSense's data "strongly" correlated with campus testing for influenza and other respiratory diseases. To be clear, researchers noted that FluSense was designed to study and predict population-based outbreaks and transmission as opposed to singular individual studies. So it won't tell you if you're sick. But it might be able to tell city administrators when a particular neighborhood is about to have an outbreak of something.

What about my privacy? — Public health detection methods shouldn't come at the expense of your privacy. Especially with the outbreak of COVID-19, multiple governments have greenlit controversial surveillance tactics in the name of protecting citizens' health. In contrast, one of the most compelling aspects of the FluSense experiment is that it relies on chunks of unidentifiable data. As researchers noted, FluSense focuses on non-speech signifiers and draws bounding boxes around thermal images in public spots, then counts the number of people in that particular diagram without distinguishing them by personal information.

Desperately-needed technology — Machine learning could benefit and expedite flu forecasting and prediction studies. If FluSense is introduced at a much more mainstream and affordable level, epidemiologists could have access to critical data and better grasp how a virus latches onto its external environment, which demographics in a particular area are most vulnerable to transmission, how high or low its virality rate is, and more.

It could also help medical experts prepare effective flu vaccines and inform governments on whether or not there is a need for travel restrictions or social distancing. Machine learning and prediction models like this can also help hospital administrators determine the need for medical supplies — something we're acutely aware of these days. We've seen how unprepared the world was for COVID-19. FluSense isn't the only thing we need to tackle viral outbreaks, but we'll take whatever help we can get.