Senior creates AI diagnostic tool, gets published in biomedical journal
November 21, 2024 at 12:39 a.m.
Before being legally permitted to get behind the wheel of a car—a test she nervously conquered in October (it ended up being easier than she thought)—Charlotte Zelin celebrated an achievement that’s difficult for academics twice her age: she got published.
And her scientific article wasn’t just published, but she appeared as the first author.
Zelin’s paper, “Rare Disease Diagnosis Using Knowledge Guided Retrieval Augmentation for ChatGPT,” was in the September issue of the Journal of Biomedical Informatics—the premier methodology journal in the field. She’s currently a Blind Brook High School senior, celebrating the publication of an article resulting from research she started as a sophomore as part of the high school’s three-year science research program.
“There is an option in the science research program to do a different study for your junior year and your senior year, or you can do one throughout the program,” Zelin said. “If you do one throughout though, you can’t present it at competitions when you’re a junior.”
Calling Zelin ambitious would be putting it lightly. While she’s receiving accolades for her first research project, she’s also currently working on a second, unrelated one that involves tracking physical, phenotypical symptoms of autism spectrum disorder (which is typically diagnosed behaviorally).
“I’m the type of person who needs to get things done,” she said. “I feel like, if I have an opportunity to do more, I’ll always do it.”
For better or for worse, “I just feel like I can do everything,” she laughed, before joking: “the secret is no sleep.”
Zelin’s journey started when she joined the science research program as a sophomore and was tasked to find a mentor to partner with through her scientific explorations. A cold email to Dr. Chunhua Weng, a professor of biomedical informatics at Columbia University, was fruitful.
“I read some of her papers and I thought they were completely interesting,” Zelin said. She and Weng ended up clicking, a relationship that is now propelling the senior toward a field she looks to pursue. “In college, I want to do biomedical engineering, it’s like the intersection of computer science and medicine. I feel like there’s so many things you can do to assist clinicians, you can build all these tools to help them. And that’s such an interesting crossover.”
And that was the ambition behind her first project, where she was able to bring forth ideas involving the hot topic of her generation: Artificial Intelligence (AI).
Using YouTube videos to teach herself how to code, Zelin created an enhanced ChatGPT model and tested its ability to theoretically help clinicians diagnose patients with rare diseases.
The rare disease diagnostic tool was trained with over 700 published case studies and research papers and programmed to give users a few diagnoses most relevant to the symptoms they disclosed. There were limitations in the information she was able to feed the ChatGPT model for enhancement because personal health records could not be used—the information would become vulnerable. Similarly, that would be a cause for concern in real-life application.
“In the future, you could have a rare disease tool like this if hospitals have a privacy agreement with Open AI,” she said, referring to the company that owns ChatGPT. “It could be integrated in electronic health records so if you were to see a patient and input all their symptoms, it could automatically do that diagnosis for them. That’s how I envision it, but there are privacy issues that need to be overcome for that.”
Zelin then tested the technology by communicating with it about theoretical patients. She’d use three different question styles, exploring prompt engineering that goes hand-in-hand with AI utilization, and sent its responses to rare disease diagnosticians from Harvard and Boston Children’s Hospital to review its answers.
“There’s a really big wave in the biomedical field called precision medicine that’s tailoring treatment specifically to an individual,” Zelin said. “Because as we’ve gone on, we’ve realized people have different ways of manifesting symptoms and everyone’s journey is so individualized, so this is kind of what my study aims to help improve on.”
In a clinical setting, rare diseases are a niche field, Zelin said—the average diagnostician has little knowledge or experience with them. Therefore, a database helping with the diagnosing process would have the potential to save patients time, money and an invaluable amount of uncertainty.
“Because the average clinician won’t know how to diagnose a rare disease, people who have a rare disease often go on this thing they call the ‘diagnostic odyssey,’” she said. “It takes (on average) five years for people to get diagnosed. It’s when patients are going from doctor to doctor, clinician to clinician, trying to figure out what’s wrong. Also, these rare diseases tend to have a heterogeneous representation or manifestation, so someone will have this symptom, and someone with the same disease will have a completely different symptom.”
When Zelin finished her project, it hadn’t even crossed her mind to try and get it published. But when her mentor Dr. Weng encouraged it, her “needs to do more” mindset couldn’t say no.
“Predominantly, I was the main driver for this study,” she said, explaining her first author status on the article. “My mentor oversaw me; I would have weekly meetings with her. But in the end, I was the one deciding on methodology. I had to do all the coding, I designed the different prompts and did the literature review at the beginning.”
In Spring on this year, she put together a manuscript, submitted her article and began the arduous peer-review process.
She described the next few months as insane, and intimidating, but also an invaluable experience. The article went through several rounds of notes, critiques and editing with professionals in the field, and Zelin had to learn how to both embrace the feedback and defend her work.
“Defending yourself was the scariest part for me,” she said. “It’s like, all these people have PhDs, they’re prominent figures in the field. I don’t want to disagree with them because they have more knowledge than me, but then I also have reasons for why I did what I did.”
When all was said and done, she slammed the laptop shut and simply waited for an answer. The email came in August.
At the time, she was staying at a friend’s house in Connecticut, crashing for a week as they both attended a week-long summer dance intensive. At that moment she knew she’d be starting her senior year of high school as a published researcher.
“It was such a casual email, but I’m sitting there being like ‘Oh, my God!’” she exclaimed. “It’s rare to see this happen for a high school student; it’s been a really cool experience.”
Comments:
You must login to comment.