Students Apply AI Skills to Pandemic Early Detection and Response

March 2018
Topics: Artificial Intelligence, Machine Learning, Disease Outbreaks, Disease Transmission, Public Health
"Early detection, early response" is the best defense against infectious disease. At a Hack2React from April 13–15, MITRE challenged students to apply their machine-learning skills to real-time disease monitoring worldwide.
Students working during the MITRE Hackathon

The specifics of the diseases that have the potential to kill thousands—or millions—of people may vary greatly. However, the best chance of combatting them is the same: early detection, early response.

By identifying an emerging pandemic and detecting how the disease is spreading, local and international healthcare organizations can mobilize earlier and take steps to halt the disease's progress.

"We believe that college students skilled in machine and deep learning can contribute new and different ideas to support health organizations' missions," says Jay Crossler, MITRE's chief engineer for Learning Systems and leader of MITRE's "Hack2React" hackathons. "When it comes to early detection, the incubation period is an inherent challenge. Infected people can pass the disease on to friends and colleagues. Or they can get on a plane and take a budding epidemic to another continent, creating a pandemic."

For a dramatic look at just how such a disease may spread, check out Dr. Larry Brilliant's TED TALK, where he outlines the plot of the 2011 movie that he consulted on, Contagion. From an apple contaminated by a bat, to the pig who eats it, to the restaurant chef who shakes hands with a character portrayed by Gwyneth Paltrow, Brilliant discusses how diseases can be transmitted and the importance of early detection, early response.

Brilliant recorded his TED TALK in 2013, but the potential for social media to play an important role is even greater now.

Applying Machine Learning to Improve Real-Time Disease Monitoring

The latest Hack2React was held at our Charlottesville, Virginia, location. We invited students from surrounding Virginia and D.C.-area colleges to strengthen their machine-learning skills to improve real-time disease monitoring. MITRE provided cloud time, machine-learning tools, and access to big data from government organizations, such as the Centers for Disease Control and Prevention. Students were invited from University of Virginia, Virginia Tech, James Madison University, George Mason University, Virginia Commonwealth University, the College of William & Mary, and other higher-learning institutions in the region.

"The Hack2React events are all about learning," Crossler says. "We supplied students with real data in the format the government uses."  

In addition, MITRE provided participants with tutorials on the technologies and data provided, plus awarded thousands of dollars in cash prizes from our internal funds.

Combining Social Media with Satellite and Other Data Sources

Michael Balazs, a technology integrator at MITRE, gave a few examples of how social media might be useful for detecting an emerging epidemic or pandemic.

"For instance, would satellite images showing an increase in the number of cars in hospital parking lots reveal an uptick in the number of people who are sick within a certain geography? What if you combined that with a search of keywords from Facebook, Twitter, and other social media sites showing that many people in the same area are not feeling well?"

Crossler also pointed out that various clues from technology and social media might reveal different things in different parts of the world. By using algorithms to track the ways in which people come in contact with one another, it may be easier to determine how diseases spread.

"About five years ago people thought the most important determinant of reducing influenza was to ensure that nurses made it to work. But we did a massive simulation model and found that within technologically strong countries like the U.S., it was just as important that the information technology folks went to work.

"Why? Because many people will stay home when they're sick if their WIFI and cell phones work—and that reduces the spread of the disease."

That idea wouldn't have much impact in a technologically underdeveloped country. But if models show that the large numbers of people who became sick also visited restaurants the week before, perhaps shutting down the restaurants and providing packets of food at home might be a good solution.

"The exciting thing about the deep-learning models we use at all our Hack2Reacts is that instead of just trying one idea at a time, the computer can try a hundred million at a time," Crossler says. "That offers the potential for ideas that no one has ever thought of before."

—by Bill Eidson


Interested in MITRE's Work?

MITRE provides affordable, effective solutions that help the government meet its most complex challenges.
Explore Job Openings

Publication Search