research

⬇ Download CV (PDF)

I study the reliability of medical image classifiers: whether a model’s accuracy rests on evidence a clinician would recognize as relevant, and how that question can be examined rather than assumed.

The work runs in two directions. Reliable medical AI models asks what a diagnostic model is actually responding to and when that response can be trusted. Clinical translation is the medical study that gives the first question something to check against. My M.S. research at Ajou University, advised by Prof. Jung-Won Lee in the Embedded & Software Lab, sits where the two meet, in ultrasound diagnosis.

Reliable Medical AI Models

A classifier can reach good diagnostic accuracy using features whose relationship to disease depends on how the images were acquired or selected rather than on the disease itself. Hospital-specific formatting, measurement markers, and acquisition settings can all carry predictive signal, and a random split preserves those relationships in training and test alike, so an evaluation can look successful while leaving the question untouched.

My interest is in the gap between a model’s stated evidence and its actual reliance. An explanation that looks anatomically reasonable is not thereby faithful to the computation, and a readout that correlates with a clinical factor has not thereby been shown to drive the prediction. I work on how those distinctions can be made measurable, and on what has to hold before a result counts as evidence rather than a plausible picture.

The same question extends to generalization. Performance that holds on internal data can fail across hospitals, devices, and patient populations, and stable aggregate accuracy can hide a change in what the model is responding to.

Clinical Translation

The clinical half is not a supporting activity. Terms like clinically meaningful evidence have no content until the medicine behind them is understood, and without that content there is nothing for an audit to check against.

So I study the diseases, the imaging findings, and the diagnostic reasoning that produce a diagnosis: which features a radiologist reads and in what order, which findings are interpreted together, what a reference standard does and does not establish, and how patients who receive one workup differ from those who receive another. I also study ultrasound acquisition itself, since operator, machine, and preset variation are a property of the examination rather than of the patient, and telling those apart is a prerequisite for calling anything a shortcut.

My study notes cover both directions in more detail, and my publications list the work that has appeared.

Selected Projects

Education

• M.S. in AI Convergence Network, Ajou University  Feb 2026 – Feb 2028 (expected) Advisor: Prof. Jung-Won Lee. Embedded & Software Lab / MIIDS Research Center. Thesis on clinical faithfulness auditing of ultrasound diagnostic models.
• B.S. in Electrical and Computer Engineering (Microdegree in Data Science & AI), Ajou University  Mar 2020 – Feb 2026 Advisor: Prof. Jung-Won Lee.

Research Experience

• Graduate Researcher, Embedded & Software Lab (ESL) / MIIDS Research Center, Ajou University  Feb 2026 – Present MSIT-funded University ICT Research Center (ITRC) Conducting M.S. research on trustworthy medical AI, including post-hoc auditing of evidence use in trained models and auditable diagnostic model design (M2, M1). Evaluating model evidence using frequency-band, attention, and attribution analyses across CNN, transformer, graph, and wavelet-based classifiers.

• Undergraduate Researcher, ITRC Project on Gallbladder Ultrasound AI  Jan 2025 – Feb 2026 MIIDS Research Center, Ajou University Contributed to an MSIT-funded project on AI-based gallbladder ultrasound diagnosis in collaboration with Ajou University Hospital; the work continued into my M.S. research. First-authored a reliability requirements study reporting a ~33% improvement in malignant-class detection sensitivity (C1). Developed clinical-knowledge-guided modeling strategies using echogenicity, texture, lesion margin, wall features, and anatomical context. Presented center research at the ITRC Talent Development Fair in 2025 and 2026.

• Graduate Researcher, Diagnostic Prediction via Robotic Motion Anomaly Detection  Dec 2025 – Nov 2026 Embedded & Software Lab (ESL), Ajou University, funded by Samsung Heavy Industries Developing machine learning models for robotic motion anomaly detection and equipment fault prediction in smart-factory systems; first-authored a monitoring-software design paper based on international reliability standards (C2).

• Undergraduate Research Intern, Dept. of Psychiatry, Ajou University Medical Center  May 2024 – Dec 2024 Participated as an external researcher in a government-funded, IRB-approved computational psychiatry project involving clinical research workflows and clinical data handling. Supervised by Prof. Taewi Kim.

• Research Assistant, Energy Center, Ajou University  Jan – Dec 2021 Operated XRD and IR spectroscopy equipment and analyzed samples for university and industry collaborators.

Teaching Experience

• Teaching Assistant (Spring 2026) → Head Teaching Assistant (Fall 2026), Logic Circuit Laboratory (undergraduate), Ajou University  Mar 2026 – Dec 2026 Supervised undergraduate digital logic labs (circuit troubleshooting, team projects, and grading) as TA in Spring 2026, with student evaluation 4.77 / 5.00 (95.4%). Served as Head TA in Fall 2026, training and coordinating the course’s teaching assistants.

• Undergraduate Research Mentor, Convergence Electronics Research, Dept. of ECE, Ajou University  Jan – Dec 2026 Mentored an undergraduate research project on ultrasound image segmentation model design and experiments.

Selected Projects & Competitions

• Self-Supervised ECG Anomaly Detection Framework, ICT Challenge 2026  Jan – Dec 2026 University ICT Research Center Program Designing, training, and optimizing lightweight self-supervised models for ECG anomaly detection.

• Explainable Diagnostic Framework for Gallbladder Polyp Classification, ICT Challenge 2025  2025 Undergraduate capstone project · Team leader · also presented at the Industry–Academia Fair Developed a clinically guided ultrasound classifier using lesion margin, contrast, and liver–gallbladder echogenicity difference as explicit model features. Achieved 93% overall accuracy and 0.90 malignant-class F1 on a public dataset.

• Conquer Health Hackathon, Medical Science Foundation Models, hosted by Lunit  Aug 2026 Team of 2 · system designer and presenter Designed Control Plane for a Frozen Clinician: an orchestration harness that controls a frozen medical foundation model (Lunit L2) with risk-adaptive clinical-evidence retrieval and source-cited answers, without retraining. Evaluated on HealthBench-based clinical dialogue tasks; presented at the final session.

Awards

• Encouragement Award, ECE Industry–Academia Fair, Ajou University  Dec 2025 Individual entry · 84 teams competed
• Social Value Award (Special Award), 2025 Capstone Design Competition, Ajou University  Nov 2025 Team leader · 1 of 12 awarded teams

Skills

• Programming. Python, PyTorch, PyTorch Geometric, Linux, Git
• Machine Learning. CNNs, vision transformers, graph neural networks, self-supervised learning
• Trustworthy AI. Shortcut analysis, attribution analysis, frequency-band analysis, model auditing
• Medical Imaging. Ultrasound imaging, clinical-factor analysis, superpixel-based image representation
• Languages. Korean (native); English (IELTS scheduled)

Professional Training

• Digital Healthcare AI Solution Development and Industry Field Experience, Center for Artificial Intelligence in Healthcare, Seoul National University Bundang Hospital  Aug 2024
• Convergence Security Workforce Training: Smart Healthcare (Basic), 21 hours, Korea Information Security Industry Association (KISIA), Ministry of Science and ICT  Jul 2024

Community Service & Military

• Community Outreach Volunteer, Sillim-dong, Seoul  Mar 2026 – Present Participate in regular community outreach, providing practical assistance and ongoing support to local residents.
• Sergeant, Republic of Korea Air Force, 15th Special Mission Wing, honorably discharged  Mar 2022 – Dec 2023

Open-Source Software

Selected repositories on GitHub:

Published papers and manuscripts in preparation are listed on the publications page.