This Portable Device Turns Coughing Sounds Into Quantifiable Health Data

UMass researchers have invented a portable surveillance device powered by machine learning can detect coughing and crowd size in real-time.

In the midst of the current COVID-19 coronavirus pandemic, it has become extremely apparent that we’re woefully unprepared for widespread medical diagnoses. Throughout this outbreak, medical professionals have struggled to quantify the spread of the disease. Part of that is due to the simple fact that we’ve had trouble actually diagnosing cases of COVID-19, but the truth is that we wouldn’t have great data even with solid diagnostic testing. That’s because even the best tests require that people go to take them. This portable device, on the other hand, is able to gather health data simply by listening to people coughing.

This device, called FluSense, was created by computer scientists from the University of Massachusetts Amherst and was tested in waiting rooms at the university’s campus clinic. The device isn’t intended to perform actual medical diagnostic tests, but rather to gain a general sense of public health. At a high level, it listens to the coughs in a room and compares that to the number of people in that room. While coughing isn’t necessarily indicative of illness — and certainly isn’t proof of a coronavirus case — it is a reasonable way to quantify data trends. A statistically significant increase in coughing per person could indicate that respiratory diseases are spreading in that area.

FluSense is built entirely from off-the-shelf parts, and could easily and affordably be implemented in waiting rooms around the world. A neural network is running on a Raspberry Pi paired with an Intel Movidius Neural Compute Stick, and has been trained to listen for coughs and watch for people. It listens through a USB microphone array, and watches through a thermal camera. The entire device fits into an enclosure that is roughly the size of a shoe box. FluSense was used to collect data from December of 2018 to July of 2019, and more than 350,000 thermal images and 21 million non-speech audio samples were gathered during that time. That data was strongly correlated with the daily illness rates at the university clinic, suggesting that FluSense would be a valuable tool for following health trends.

Related articles
Sponsored articles
Related articles