How ECGR 5116 Shifted My Perspective on AI in Medicine

Student Submission written by Amirhossein Ghasemi

AI for Biomedical Applications, Medical Image Generation

Taking ECGR 5116 was a great experience that helped me refine my understanding of AI in medicine. Coming into the course, I already had some background in applying AI to ophthalmology, but this class gave me a more structured way to think about the challenges and real-world implications of the models I work with.

One of the things I appreciated most about the course was how well it balanced theory with practical application. The lectures weren’t just about throwing concepts at us—they were engaging and always tied back to real medical AI problems. We also had to present AI papers in medicine, which led to insightful discussions on different aspects of AI models, including their strengths and limitations.

A Challenging but Rewarding Project

For my final project, I worked on text-to-image generation for OCT scans. I had worked on AI models that interpret medical textual data before and not imaging data, so this was a new challenge. Medical language is highly specific—subtle changes in wording can imply different conditions or severities.

At first, my model’s outputs were inconsistent. Sometimes the generated OCT scans made sense; other times, they were completely off. I had to rethink how the textual descriptions were structured and refine the training process to capture more nuanced details. Through experiments, I improved the outputs, though it was clear that generating clinically accurate images required more than just technical adjustments—it required a deep understanding of how medical professionals interpret and describe images.

A Course That Made Me Think Differently

What I appreciated most about ECGR 5116 was that it wasn’t just about coding or optimizing models. The discussions pushed us to think critically about AI’s role in medicine. The instructor encouraged questions and discussions on state-of-the-art (SOTA) models, which made complex topics more engaging and thought-provoking.

By the end of the course, I walked away with more than just technical knowledge. I had a better understanding of the responsibility that comes with building AI for healthcare. It’s not just about developing high-performing models—it’s about ensuring they are reliable, unbiased, and actually beneficial in real-world medical applications.

This course reshaped how I approach my own research. Now, when I work on AI models for ophthalmology, I don’t just think about performance metrics—I think about whether these models can truly help doctors and patients. ECGR 5116 made me more aware of the broader impact of AI in medicine, and for that, I’m grateful