Featured Researcher: Nathan Hadjiyski '25 : About Student Research : Office of Undergraduate Research : University of Rochester

Featured Researcher: Nathan Hadjiyski '25

About the Student Researcher

Major(s) and minor(s)

B.S. in Computer Science

Hometown - city, state, country?

Ann Arbor, Michigan

Area of interest

AI/ML Medical Imaging Research

Types of research positions

What's your research story?

Since Spring 2022, I've been fortunate to work as a research assistant in Professor Axel Wismüller's lab at the University of Rochester, contributing to deep learning research in medical imaging. My work has explored cross-modal representation learning, attempting to find new ways of connecting radiology reports with chest X-ray data to potentially improve diagnostic processes. With the guidance of my mentor and collaborators, I've been learning about developing transformer-based models and working on metrics to assess diagnostic accuracy. These research efforts have resulted in first-author publications and presentations at SPIE conferences, where I've shared my work through both poster and oral presentations.

Being a part of a graduate-level research environment as an undergraduate has been transformative, offering unprecedented opportunities to explore cutting-edge AI methodologies. This experience has cultivated my ability to think creatively, challenge existing paradigms, and develop innovative solutions to complex interdisciplinary challenges. The collaborative nature of my research has profoundly deepened my understanding of how sophisticated AI technologies can drive meaningful advancements in healthcare.

In Summer 2023, I interned at the FDA as an ORISE Fellow, addressing challenges in medical AI systems such as model drift, bias quantification, and data harmonization. I developed tools to process large datasets, optimized workflows, and enhanced the reliability of AI models. This experience provided invaluable insights into navigating the complexities of real-world medical AI applications, deepening my understanding of regulatory considerations and the importance of model stability in healthcare contexts.

The following summer, I joined Optum (UnitedHealth Group) as a Software Engineering Intern, where I built an AI-driven phone assistant leveraging LLMs and modular frameworks to improve provider database management. This role allowed me to gain hands-on experience in developing scalable systems that integrate AI with practical healthcare solutions.

My interests continue to evolve, and I'm curious about exploring areas like video understanding and multimodal AI. I'm excited to join Optum as a Software Engineer in July 2025, hoping to contribute meaningfully to healthcare technology. I also will be working towards completing my M.S. in Computer Science through the GEAR (Graduate Engineering at Rochester) program at the University of Rochester.

Outside of work, I enjoy hiking, visiting museums, and playing tennis.

I'm grateful for the opportunities I've had and look forward to continuing to learn and grow! :)

—Academic and professional connections: —

• Google Scholar: https://scholar.google.com/citations?user=H5-56OMAAAAJ&hl=en

• SPIE Profile: https://spie.org/profile/Nathan.Hadjiyski-4413786

• LinkedIn: https://www.linkedin.com/in/nathan-hadjiyski/

How did you initially secure your research position?

During high school, I taught myself to use the TensorFlow library and applied the Inception V3 deep learning model to classify different stages of kidney cancer through transfer learning under the guidance of a urologist. This project introduced me to foundational concepts in data collection, preprocessing, and the application of deep learning to solve medical challenges. The outcomes of this work included two conference presentations: one at IEEE (“Kidney Cancer Staging: Deep Learning Neural Network-Based Approach”) and another at the Tech Connect Innovation Conference (“Kidney Cancer Staging Using Deep Learning Neural Networks: Comparing Models Trained on the Whole Kidney with Cancer and Only the Cancer”). I was also honored to receive the South East Michigan 2019 IEEE Spring Section Conference Best Poster Award for this project.

This early experience solidified my passion for research at the intersection of AI and healthcare. As I transitioned to university, I explored the faculty directory to find mentors whose research aligned with my interests. I was particularly drawn to Dr. Axel W. E. Wismüller’s lab due to his expertise in both AI and radiology, as well as his emphasis on bridging the gap between cutting-edge research and clinical practice.

I recognized that Dr. Wismüller’s lab was a graduate-level research environment, which could be an additional challenge for an undergraduate student. However, I saw this as an exciting opportunity to prove my value and grow as a researcher. I prepared thoroughly for my application and interview, ensuring I could articulate my prior experiences, demonstrate a clear understanding of my work, and show how I could contribute to the lab’s goals.

During the summer of 2022, I collaborated closely with PhD student Ali Vosoughi to develop a cutting-edge machine learning model. We explored integrating cross-modal learning techniques to combine radiology reports and X-ray chest images, expanding the capabilities of existing methods. Our work led to the publication of my first-authored paper in the lab, “Cross-Modal Global-Local Representation Learning from Radiology Reports and X-Ray Chest Images,” which I presented at the 2023 SPIE Medical Imaging Conference.

For my next first-authored publication in the lab, I worked closely with PhD student Akhil Kasturi on another impactful project. Together, we applied a memory-driven transformer architecture to improve radiology report generation from chest X-rays, establishing a new quantitative metric for evaluating model performance. This resulted in the publication of “Leveraging a Memory-Driven Transformer for Efficient Radiology Report Generation from Chest X-Rays to Establish a Quantitative Metric,” which I presented at the 2024 SPIE Emerging Topics in Artificial Intelligence (ETAI) Conference.

Working in a graduate lab has been a rewarding challenge, and I am deeply grateful for the mentorship and collaborative environment that have enabled me to contribute meaningfully to ongoing projects. This experience has reinforced my commitment to advancing AI applications in healthcare and has been instrumental in shaping my academic and professional aspirations.

Departments/programs of research

• University of Rochester Medical Center, Wismüller Lab: Deep learning and medical image analysis.

• FDA CDRH: Regulatory-focused AI model reliability and data harmonization.

• Optum (UnitedHealth Group): Real-world applications of AI in healthcare technology.

Has your research experience enabled you to qualify for/apply to other awards or scholarships?

• Honorable Mention Award at 2024 SPIE Medical Imaging Conference - Issued February 2024

• Accepted in the competitive Graduate Engineering at Rochester (GEAR) 5 year Master's Program • Recipient of Research and Innovation Grant (RIG) - funded by the University of Rochester for independent research

• Recipient of the Whipple Science & Research Scholarship

• South East Michigan 2019 IEEE Spring Section Conference Best Poster Award – Issued May 2019

Any research presentations, awards, or publications?

• Nathan Hadjiyski, Ali Vosoughi, Axel Wismüller, "Cross modal global local representation learning from radiology reports and x-ray chest images," Proc. SPIE 12465, Medical Imaging 2023: Computer-Aided Diagnosis. Poster Presentation.

• Nathan Hadjiyski, Akhil Kasturi, Ali Vosoughi, Axel Wismüller, "Leveraging a memory-driven transformer for efficient radiology report generation from chest x-rays to establish a quantitative metric," Proc. SPIE 13118, Emerging Topics in Artificial Intelligence (ETAI) 2024. Oral Presentation.

Links - Any papers you published from your research, your work team's page, previous articles written about your research, social media links to tag for your project, etc

—Research Pre-College:—

• Nathan Hadjiyski, "Kidney Cancer Staging: Deep Learning Neural Network Based Approach," 2020 International Conference on e-Health and Bioengineering (EHB), Iasi, Romania, 2020, pp. 1-4, doi: 10.1109/EHB50910.2020.9280188. Oral presentation.

• Nathan Hadjiyski, Kidney Cancer Staging using Deep Learning Neural Network: Comparing Models Trained on Whole Kidney with Cancer and Only the Cancer, 2021 TechConnect World Innovation Conference and Expo, Oct 18, 2021, Washington, DC; Proc. TechConnect Briefs 2021, pp. 161 – 164, ISBN: 978-0-578-99550-2, Oral presentation.

—Research @ Wismüller Lab:—

• Akhil Kasturi, Ali Vosoughi, Nathan Hadjiyski, Larry Stockmaster, William J. Sehnert, Axel Wismüller, "Anatomical landmark detection in chest x-ray images using transformer-based networks," Proc. SPIE 12927, Medical Imaging 2024: Computer-Aided Diagnosis, 129272Q (3 April 2024); https://doi.org/10.1117/12.3006881. Poster Presentation **Received Honorable Mention Award**.

—Research @ FDA CDRH:—

Smriti Prathapan, Ravi K. Samala, Nathan Hadjiyski, Pierre-François D’Haese, Fabien Maldonado, Phuong Nguyen, Yelena Yesha, Berkman Sahiner, "Quantifying input data drift in medical machine learning models by detecting change-points in time-series data," Proc. SPIE 12927, Medical Imaging 2024: Computer-Aided Diagnosis, 129270E (3 April 2024); https://doi.org/10.1117/12.3008771. Oral presentation.

Can you share some "lessons learned" as a result of your undergraduate research experience?

Through my research experiences at the University of Rochester, I have come to understand that structure and collaboration are the foundational pillars of achieving impactful results. Establishing clear goals, timelines, and project scopes from the outset has proven to be transform

ative. By breaking down complex projects into manageable tasks, I’ve learned to approach even the most daunting challenges with clarity and composure, effectively minimizing stress and ensuring continuous progress toward deadlines.

Equally important is the power of communication and collaboration. Regular check-ins with mentors and collaborators are not only critical checkpoints but also invaluable opportunities for growth. These interactions offer fresh perspectives that help guard against tunnel vision and refine the project’s trajectory. They often reveal the broader significance of the work, uncovering new avenues for discovery and enhancing the novelty of the approach. It is through the synergy of collaboration that our lab’s most innovative and groundbreaking ideas come to fruition.

What advice can you share with new undergraduate researchers?

• Ask for Help Confidently: Don’t hesitate to reach out when you encounter challenges. Seeking guidance from mentors or peers can save you significant time and energy. That said, it’s essential to approach these moments thoughtfully. When asking for help, prepare a clear summary of your attempts to solve the problem and the specific questions you need answered. This not only demonstrates initiative but also helps others provide more targeted and effective support.

• Dream Big and Stay Curious: Never let the scale or complexity of a project deter you from pursuing it. Even if something feels beyond your reach, there’s immense value in exploring your curiosity and working creatively within your circumstances. The process of tackling ambitious ideas often leads to unexpected breakthroughs and personal growth, even if the final result differs from the initial vision.

• Embrace Adaptability: Research rarely goes exactly as planned, and unexpected challenges are inevitable. When things deviate from your expectations, take a step back, assess the situation calmly, and reframe the problem as an opportunity to learn and grow. Staying composed under pressure allows you to make rational, productive decisions that lead to better outcomes and reduce frustration.