Portable AI Device Used to Gather Health Data for Predicting Pandemics

The device, named FluSense, can detect coughing and crowd size in real-time. Data collected can be analyzed for the monitoring of flu-like illnesses and flu trends. 

Coughing could help health experts better understand the distribution and transmission of respiratory outbreaks in crowds. That is according to the University of Massachusetts Amherst (UMass Amherst) researchers who have created an artificial intelligence (AI) flu monitoring device called FluSense. 

The UMass Amherst researchers say that their new edge computing platform could be used in healthcare settings like hospitals and waiting rooms, in addition to larger public spaces, to increase the number of health surveillance tools at the disposal of health experts. It could also be used to help forecast seasonal flu and other respiratory outbreaks, news that some will undoubtedly find most bittersweet amidst the current COVID-19 pandemic. 

The device makes use of a microphone array, thermal camera, and neural computing engine on the edge to passively monitor and characterize speech and coughing sounds alongside changes in crowd density. Despite being equipped with a microphone and camera, no personal information is recorded or stored. 

 

Creating the Device

Before building the device, the researchers had to develop a lab-based cough model. They then trained a deep neural network classifier to draw bounding boxes on thermal images representing people. “Our main goal was to build predictive models at the population level, not the individual level,” professor Tauhidur Rahman said.

When they had a working device, the researchers focused on four waiting rooms at UMass Amherst’s health services clinic by deploying the FluSense device in them. Between December 2018 and July 2019, the researchers collected and studied over 350,000 thermal images and 21 million non-speech audio clips. 

 

The components found in the FluSense device. Image credit: UMass Amherst

 

Using data collected during this time, FluSense was able to predict illness rates with a high rate of accuracy. The researchers noted that captured data “strongly” correlated with campus testing results for influenza and other respiratory diseases, with a Pearson correlation coefficient of 0.95. This demonstrates the value to be found in combining AI with edge computing to enable the gathering and analysis of data right at the source. 

 

Predicting Population-Based Outbreaks

It is important to note that FluSense is not designed to identify when individuals are or may be ill. Instead, it is designed to study and predict population-based outbreaks and transmission. In this way, it could be used to help public health authorities and city administrators identify when an outbreak is or could be on the horizon, and where. 

FluSense could also be used to help support current influenza prediction efforts such as the FluSight Network—a consortium of flu forecasting teams. Rahman went on to add, “I thought if we could capture coughing or sneezing sounds from public spaces where a lot of people naturally congregate, we could utilize this information as a new source of data for predicting epidemiologic trends,” 

The next step for FluSense is for it to be tested in other public areas and geographic locations, “…we want to validate it beyond this specific hospital setting and show that we can generalize across locations,” said professor Andrew Lover. 

 

A Much-Needed Technology

FluSense demonstrates that machine learning can benefit and expedite flu forecasting and prediction. If introduced at a more mainstream level, epidemiologists and other health experts could access critical data and gain a better understanding of how a virus spreads and identify external influences, vulnerable demographics, virality rate, and more.

It could also help medical experts, governments, and even industry prepare for flu outbreaks more effectively by helping identify whether there is a need for travel restrictions, social distancing or medical supplies.