AI and Edge Computing Combine in Portable Platform for Flu and Potentially Coronavirus Pandemic Forecasting

The FluSense device houses these components. [UMass Amherst]

Scientists at the University of Massachusetts (UMass), Amherst, have developed a portable surveillance device that can directly monitor influenza-like illnesses (ILI) and flu trends through the use of machine learning models that analyze data on coughing sounds and crowd size, caught in real time, in environments such as healthcare waiting rooms. The creators say the edge-computing platform, FluSense, which they could see used in areas such as hospitals or larger public spaces, could expand the suite of health surveillance tools that are currently used to forecast seasonal flu, but potentially also other viral respiratory outbreaks, such as the COVID-19 pandemic or SARS.

Such models can help to save lives by directly informing public health responses during a flu epidemic. The data sources might also help to establish the timing for flu vaccine campaigns, potential travel restrictions, and the allocation of medical supplies. “This may allow us to predict flu trends in a much more accurate manner,” said Tauhidur Rahman, PhD, assistant professor of computer and information sciences, and Forsad Al Hossain, PhD student and graduate research assistant, who reported on the FluSense platform in The Proceedings of the Association for Computing Machinery on Interactive, Mobile, Wearable and Ubiquitous Technologies, in a paper titled, “FluSense: A Contactless Syndromic Surveillance Platform for Influenza-Like Illness in Hospital Waiting Areas.” Co-author Rahman is adviser to Al Hossain, who is lead author of the report.

The authors cite Centers for Disease Control and Prevention (CDC) figures, which estimate that flu infections lead to 4,000–23,000 medical visits, and 12,000–79,000 deaths each season. “The estimated annual economic impact of these infections is $47 and $150 billion in the United States alone,” they wrote. The CDC currently aggregates reports on outpatient influenza-like illness from state-level public health agencies, and monitors flu-related hospitalization and death rates, as part of its method for monitoring the burden of flu during the flu season. Virologic surveillance collected by clinical laboratories affiliated with the CDC detailed information about the relative prevalence of influenza subtypes at a given timepoint. However, the FluSense developers noted, current surveillance systems do have limitations, including, “most notably,” the substantial time lag in data reporting.

More recent research has looked at incorporating nontraditional data sources into infectious disease forecasting, or in an attempt to address the limitations of existing systems. Such “digital epidemiology,” sources might include social media, internet search trends, satellite imagery, climate data, and smartphone data. However, some of these forms of data may only be available in a particular context, or may be unavailable or unreliable in resource-limited settings. Moreover, the authors noted, “Many of these nontraditional data sources have shown some promise but are inherently limited in that they do not directly measure biological signals of infection or associated physical symptoms … Currently, no automatic or readily-scalable methods exist for real-time measurement of these physical symptoms of ILI directly from a crowd.”

The FluSense platform developed by Rhaman and colleagues combines a low-cost microphone array and thermal imaging data with a Raspberry Pi and neural computing engine. The system passively and continuously characterizes speech and cough sounds along with changes in crowd density, in a real-time manner, the scientists explained. Importantly, no personally identifiable information—such as speech data or distinguishing images—is stored.

Computer scientists in Rahman’s Mosaic Lab develop sensors to observe human health and behavior. For the FluSense development, the researchers first developed a lab-based cough model. Then they trained the deep neural network classifier to draw bounding boxes on thermal images representing people, and then to count them. “Our main goal was to build predictive models at the population level, not the individual level,” Rahman commented.

Tauhidur Rahman, PhD, left, and Forsad Al Hossain display their FluSense device. [UMass Amherst]

To give their invention a real-world tryout, the FluSense inventors partnered with George Corey, PhD, executive director of University Health Services; biostatistician Nicholas Reich, PhD, director of the UMass-based CDC Influenza Forecasting Center of Excellence; and epidemiologist Andrew Lover, PhD, a vector-borne disease expert and assistant professor in the School of Public Health and Health Sciences.

As an initial investigation, the researchers placed FluSense devices, encased in a rectangular box about the size of a large dictionary, in four healthcare waiting rooms at UMass’s University Health Services clinic. From December 2018 to July 2019, the FluSense devices collected and analyzed more than 350,000 thermal images and 21 million non-speech audio samples from the public waiting areas.

“I’ve been interested in non-speech body sounds for a long time,” Rahman said. “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 researchers found that FluSense was accurately able to predict daily illness rates at the university clinic. Multiple and complementary sets of FluSense signals “strongly correlated” with laboratory-based testing for flu-like illnesses and influenza itself. “To the best of our knowledge, this is the first work that establishes the link between cough sensing and influenza-like-illness trends at the scale of a college/university community,” they wrote. “Our results suggest that multiple and complementary sets of FluSense signals exhibited strong correlations with both ILI counts and laboratory-confirmed influenza infections on the same campus.”

The team projects that the early symptom-related information captured by FluSense could provide valuable additional and complementary information to current influenza prediction efforts, such as the FluSight Network, which is a multidisciplinary consortium of flu forecasting teams, including the Reich Lab at UMass Amherst. “We show that the total daily cough counts exhibited strong correlations with laboratory-confirmed influenza infections on campus,” the scientists wrote. “Additionally, the thermal camera images paired with a neural network model was able to accurately estimate the total number of patients seen at the clinic each day, which was then used to quantify incidence rates (e.g., total cough count per crowd size) that is highly informative of the daily ILI and confirmed influenza case counts.”

The researchers concluded that their study demonstrates the feasibility of using the edge-computing platform to capture cough and underlying population counts in a noisy environment, anonymously, to provide important epidemiological data about influenza trends. “This validates the premise of the FluSense platform for routine public health surveillance,” they wrote. “… these findings illustrate one way this type of edge-computing sensor platform can be used to improve the timeliness and effectiveness of current influenza-like illness prediction models.”

Al Hossain maintains that FluSense also exemplifies the power of combining artificial intelligence with edge computing, which enables data to be gathered and analyzed right at the data’s source. “We are trying to bring machine-learning systems to the edge,” he said, referring to the compact components inside the FluSense device. “All of the processing happens right here. These systems are becoming cheaper and more powerful.”

The next step would be to test FluSense in other public areas and geographic locations. “We have the initial validation that the coughing indeed has a correlation with influenza-related illness,” Lover says. “Now we want to validate it beyond this specific hospital setting and show that we can generalize across locations.”