AI Tools Advance Study and Treatment of Hearing Loss

Christopher Buswinka has always loved music, from listening to records with his father on their Hi-Fi sound system to playing piano at home and trumpet in the school band.

Studying bioengineering applications for hearing loss as an undergraduate at the University of Michigan, including the biocompatibility of cochlear implants, offered a way to combine his interests and technical aspirations.

“As you get to learn more about the hearing system, you come to appreciate the biology of it, the complexity of it,” he said. “It’s really cool to see how the biology can interpret the sound that you like to listen to.”

Buswinka was accepted into the Harvard Kenneth C. Griffin Graduate School of Arts and Sciences as a PhD student in the Speech and Hearing Bioscience and Technology (SHBT) program in 2018. There he joined the Harvard Medical School lab of Artur Indzhykulian, expecting to specialize in applying electron microscopy to cochlear research.

So, his current focus on artificial intelligence came as kind of a surprise.

Even though he took a class on AI and statistics early in his coursework, that wasn’t the turning point for him. Instead, the disruptions caused by the COVID-19 pandemic altered his plans. He moved home and, like so many others, was at a loss.

“We all got shut down,” he said. “You can’t take fancy photos with expensive microscopes when you’re in your parents’ basement.”

But he soon realized he could contribute even if he wasn’t standing at a lab bench.

“There’s a beautiful amount of data that we collect in our lab and as a field. Most of it is imaging, and most of that we’re not using because we can’t analyze it properly,” he said. “I thought, ‘Well, let me give it a try.’”

Indzhykulian bought and sent him computer parts, which Buswinka assembled where he lived and started using AI-based techniques to work through the lab’s backlog of unsorted data.

The effort started him down a path to meet a technical need in the science of hearing, enabling deeper data analysis that could in turn drive the discovery of new treatments for hearing loss.

“There have not been many easy-to-use AI systems for the study of the inner ear,” he said. “The tools I’ve developed are helping lay the foundation for new advances in our realm of biology.”

He added, “I never intended to do this when I entered grad school, but it was a happy accident and I have really enjoyed working on it thus far, even though I’m kind of not a biologist anymore.”

AI and the inner ear

Buswinka’s thesis, which he defended in September 2024, focused on applications of machine learning to the study of the inner ear. He developed two tools he’s particularly proud of.

One is a program to help researchers make the most of the reams of data they get when imaging the hair cells of the cochlea, which can number between 2,500 and 4,000 cells, depending on the animal. Researchers often make estimates rather than counting each cell manually, but that could leave interesting information undocumented.

Buswinka addressed this need by building an algorithm-powered software informed by fully analyzed and labeled images from the Indzhykulian Lab and those from colleagues throughout the field.

“Hopefully as adoption, which has been growing, gets larger, the new standard in the field will be to use the models that I have created to use all of the data they collect and hopefully do better science,” he said.

In a more niche project, he developed a tool to analyze the health of mitochondria in cochlear hair cells. These cells don’t regenerate, so what you are born with is all nature is going to give you, Buswinka said. He and members of his lab asked: Do loud noises destroy these cells by damaging their mitochondria?

At the time, most deep-learning models for analyzing mitochondrial images were optimized for neural tissue and did not work well to answer Buswinka’s question. So he created an algorithm specifically to handle the large, high-resolution, three-dimensional electron microscope images of the mitochondria-packed hair cells.

“It worked really, really well,” he said. “It allowed us to probe some of the biological effects of noise or different drugs in these cells.”

Expanding the research to other students and sensory systems

With AI now a major topic in public discourse, Buswinka notes that there is a world of machine learning that has been aiding researchers in specific situations for years separate from the more recent generative AI programs powering tools such as ChatGPT.

“Rather than solving everyone’s problem at the same time, the AI tools I’ve created have tackled a specific problem that was very challenging to solve in any other way,” he said.

Buswinka started paying forward the mentorship and AI education he received at HMS. He enlisted high school and undergraduate interns to annotate the data used to train his models and took time to explain how AI works and show them why their work was important. He made sure they all received credit for contributing to the papers and projects.

Buswinka has now taken his interest in AI and the senses to a computer vision start-up on Boston’s North Shore. He lives there with his girlfriend and his cat Sydney (and fondly remembers his cat Darwin, who died around the time of Buswinka’s defense), to whom he credits much of his success.

“I believe computer vision and machine learning have great potential to enable better science,” he said.

Whether Buswinka next pursues entrepreneurship, an academic position, or another career path, he believes his work will be useful.

“Science is a noble goal and should be pursued, especially when the outcomes of our work can benefit many people,” he said.

“Ultimately, I think Harvard was the best place for me to do the work that I wanted to do. SHBT was phenomenal,” he added. “They were just lovely people to be around. Extremely smart, very supportive. I’d do it again if I could.”

Article originally appeared on Harvard Medical School