As an international student, Michael Xu ’25 is required to get a tuberculin skin test before returning to campus each year. Two summers ago, Michael tested positive—it was a false positive. The experience left Michael wondering, could there be a more efficient, less intrusive method of detecting tuberculosis (TB)?

“After some research, I found that there are acoustical differences between patients with respiratory diseases that even trained physicians struggle to distinguish,” Michael explains. “But with the proliferation of Artificial Intelligence (AI) and crowdsourced datasets, there is the possibility for Machine Learning (ML) algorithms to listen to millions of audio clips and develop a more nuanced understanding of sound.”

And that, Michael thought, could mean that a smartphone app could be the answer he was looking for. Not long after his false positive test, Michael began taking online ML courses.

“Like many others, I was amazed by the potential of AI after ChatGPT-3 came out,” says Michael. “Since I had a foundation in Python from taking Honors Computer Science at Taft, I quickly became accustomed to the research world, reading research and code from leading AI developers.”

He found ML models built from on the largest TB benchmark dataset, CODA (COugh Diagnostic Algorithm). He also met Faisal: both were rising juniors at the time, and both had a strong interest in ML. They decided to continue Michael’s TB research.

“We sourced code from the top performing models and incorporated novel Machine Learning techniques, including Audio MAE from Facebook Research and Two-Stage Training by OpenAI,” Michael explains. “Our idea is to democratize TB detection by diagnosing it using just the sound of a patient's cough. Our model achieved an AUROC (accuracy) of 0.843 when evaluated on the largest benchmark TB cough audio dataset, outperforming all other published techniques to date.”

Michael and Faisal took their research to a regional science fair near Faisal’s home in California; the event happened to coincide with Michael’s Taft spring break. The strength of their concept and research earned Michael and Faisal a spot in the 2024 Regeneron International Science and Engineering Fair (ISEF), the world’s largest pre-college STEM competition. They earned the United States Agency for International Development (USAID) First Place Award for Global Health, which came with a $5,000 prize. They also earned Fourth Place honors in the Robotics and Intelligent Machines category, and an additional $500.

“Our work has the potential to save millions of lives in lower-and-middle income countries, putting an end to increasingly exacerbated health disparities,” notes Michael. “Instead of walking many kilometers to a clinic, people can download an app on their phone and cough into their microphones. Because TB is easily treated when diagnosed early, increasing TB awareness with accessible diagnosis is crucial to eradicating the disease.”

Learn more about Michael and Faisal’s work through their Regeneron ISEF virtual booth.