Sounds Like

For computer science students with a passion for solving health problems, the COVID-19 pandemic served as an opportunity to pursue a data research project aimed at improving a globally significant health crisis. Seniors Shryans Goyal and Will Mundy, both computer science majors, teamed up to explore a novel approach to COVID-19 diagnosis by analyzing audio samples of patients’ coughs. Their project, CoughNet, won the team award at the Rice Undergraduate Research Symposium (RURS), an event for undergraduates across all disciplines to present their research projects and receive feedback and recognition. 

“As Rice students, we have been so fortunate to have a plethora of testing available on campus, and that’s also true for many across the country,” Goyal said. “But many parts of the world don’t have that degree of access to continuous testing — the most effective tool in controlling the virus. That’s why we sought a way to incorporate deep learning to detect COVID-19 through coughs.” The duo aimed to create a prescreener of sorts that could indicate whether or not a person should pursue further COVID-19 testing by analyzing a cough sample.

Goyal and Mundy proposed a novel use of WaveGAN, an audio network, to generate synthetic COVID-positive and COVID-negative cough recordings as a form of data augmentation. “While our system was able to produce a high degree of accuracy, we weren’t able to reach the same levels that PCR [polymerase chain reaction] or rapid tests indicate,” Mundy said. “We were happy to find that the results improved as we gathered more data, which is not always the case.”

Goyal and Mundy believe further data augmentation will continue to improve cough accuracy. “We also believe there is tremendous opportunity to leverage WaveGAN and cough samples to distinguish between other respiratory-related diseases and illnesses,” Goyal said. “We were limited by the amount of data we had, but we’re confident that this is a promising avenue for continued research.” CoughNet has inspired them both to continue exploring deep learning for health care applications after graduation. 

— Kendall Hebert