The future of heart disease detection and prevention may have started with an opera singer.
When Joshua Hutcheson heard about how his wife — then a grad student — was studying how subtle changes in vocal chords could affect a singer's voice, he started thinking about the heart in a completely new way.
He began to wonder: If trained ears can pick up imperceptible differences in voice, could artificial intelligence detect subtle changes in the sound of a heartbeat?
At Florida International University, Hutcheson is the director at the Center for Innovation in Cardiovascular Health, where he leads a lab of students and researchers.
He began this line of research in 2018, aiming to better understand why cardiovascular diseases develop and how to better detect them sooner.
The team's latest diagnostic approach combines digital stethoscopes and machine learning to analyze heartbeats, similar to how a musician may pick up an off-key note.
According to the Centers for Disease Control and Prevention, heart disease remains the leading cause of death globally, often without symptoms until a serious event occurs.
A new way to listen
With heart sounds collected from the digital stethoscopes, researchers feed that audio data into a computer for further analysis.
"This would just give us another tool. … Then your doctor can work with you to then make subtle changes or put you on a treatment that really gives you a better quality of life moving forward rather than, say, waiting until you've had a major heart attack," Hutcheson said.

The goal is to train the machine learning or AI model to detect subtle changes in heartbeats that human ears may miss, offering the potential for earlier, preventative care.
" What we're doing is taking that signal and using machine learning to try to analyze that," Hutcheson explains. "To pick out very subtle changes that indicate the presence of disease."
Health and AI intersect
To train the algorithm, the team began with pre-clinical data from lab models, including collecting sounds from healthy and diseased mice hearts, with some data from human hearts as well.
The sounds capture tiny mechanical vibrations, sometimes caused by stiffening arteries or narrowing heart vessels, that can indicate the early onset of disease.
"We aren't telling the computer what it is," Hutcheson explains. "Then we look at how well it's able to then predict whether it's diseased or not — and we see about 95% accuracy on it."
It means the model has the potential to catch sounds that doctors at a regular checkup sometimes miss, he explained.
" There [are] studies out there suggesting that a doctor listening through a stethoscope can pick up the initial stages of disease only about 30 to 40% of the time, so that's a major improvement," Hutcheson said.
The model is being trained to recognize acoustic signatures of calcification— a condition where soft tissue in the arteries hardens into bone-like material, often with no previous symptoms.
Hutcheson says the goal is to collect enough data and evidence to eventually move the algorithm into clinical use. That could mean a low-cost tool that patients use at home, transmitting data directly to their physician.

"We think it could be something like a wearable blood pressure monitor," Hutcheson said. "This could enable someone at home to take these measurements, have them transmitted to a system their doctor can access and then regularly monitor it, whether it's for diagnosis or for ongoing treatment."
That kind of access could have a real impact, Hutcheson says, especially for patients in communities where regular access to cardiologists or advanced imaging tools can be limited.
"Anyone who doesn't have adequate access," Hutcheson said. "Whether that be in a rural community, or cannot afford regular access to health care, we think it can have a tremendous impact there."
Beyond just cardiology
The research is supported by grants from the Florida Heart Research Foundation and Baptist Health Hospital in Kendall, and includes collaborations with institutions like the University of Miami and other clinical partners.
While heart health is the current focus, Hutcheson believes this sound-based approach has a potential for growth and could extend well beyond cardiology.
" Anything that's producing sound that could become abnormal with changes, in the tissue, in the body, this could be used for, so I can certainly imagine bone health, muscle health, anything like that," he says. "I don't see why it couldn't be used."
Hutcheson says this convergence of sound science, AI and health care has the potential to open many new doors.
"I think that's really where the magic happens," Hutcheson said. "When you can get these people together… working on a common problem and providing expertise across the board."
Copyright 2025 WLRN Public Media