research
Trustworthy medical AI grounded in clinical evidence.
Trustworthy Medical AI Grounded in Clinical Evidence
My research asks a simple but consequential question:
When a medical AI model is accurate, is it accurate for clinically valid reasons?
High predictive performance alone does not guarantee that a model relies on evidence clinicians would consider meaningful. I study how to audit model reasoning, identify shortcut reliance, and evaluate whether interpretable model readouts remain faithful to independent clinical factors.
Clinical Faithfulness
My primary research focuses on clinical faithfulness: whether a model’s interpretable evidence meaningfully aligns with clinical severity, diagnostic factors, or other independent clinical information.
I am particularly interested in post-hoc auditing methods that can evaluate model reasoning without retraining the model or requiring direct faithfulness annotations.
Ultrasound AI
I study ultrasound AI that reflects clinically and physically meaningful image characteristics — echogenicity, texture, lesion boundaries, anatomical context, and frequency structure.
Rather than treating interpretability as an additional visualization step, I aim to connect computational evidence with the factors clinicians actually use in practice.
Demonstrating a gallbladder ultrasound AI system — diagnostic readout with a probability heatmap and edge-based evidence.
Research Interests
- Clinical faithfulness and model auditing
- Shortcut learning and spurious correlations
- Trustworthy and interpretable medical AI
- Ultrasound image analysis
- Frequency-based and graph-based representations
- Reliability monitoring for deployed AI systems
Selected Projects
Clinical Faithfulness Auditing for Medical Image Classifiers
Developing post-hoc methods to examine whether interpretable model readouts rely on clinically valid evidence rather than shortcut features — without retraining the model or requiring faithfulness annotations.
Gallbladder Ultrasound AI
Research on trustworthy gallbladder ultrasound classification in collaboration with Ajou University Hospital, including reliability requirements analysis and clinically grounded representation learning.
→ ACK 2025 (first author)Predictive Maintenance Monitoring
Designed fault-prediction models and reliability monitoring software for industrial equipment in a smart-factory research project with Samsung Heavy Industries, grounded in international reliability standards.
→ KCSE 2026 (first author)Education
- M.S., AI Convergence Network, Ajou University Feb 2026 – Feb 2028 (expected) Embedded & Software Lab / MIIDS Research Center. Thesis on clinical faithfulness auditing of ultrasound diagnostic models.
- B.S., Electrical and Computer Engineering (Microdegree in Data Science & AI), Ajou University Mar 2020 – Feb 2026
Research Experience
- Graduate Researcher, Embedded & Software Lab, Ajou University 2026 – Present Clinical faithfulness auditing and shortcut analysis of ultrasound diagnostic models (M.S. thesis).
- Undergraduate Researcher, ITRC Gallbladder Ultrasound AI Project (with Ajou University Hospital) 2025 – 2026 Reliability requirements analysis and clinically grounded modeling; first-authored ACK 2025.
- Research Intern, Samsung Heavy Industries Project, Embedded & Software Lab 2025 – 2026 Fault-prediction models and reliability monitoring software; first-authored KCSE 2026.
- Research Intern, Dept. of Psychiatry, Ajou University Hospital 2024 Government-funded computational psychiatry project; IRB-approved clinical research workflows.
Teaching
- Teaching Assistant, Logic Circuit Experiment (논리회로실험), Dept. of Electrical and Computer Engineering, Ajou University Spring 2026 Led weekly digital-logic lab sessions and guided students through circuit implementation. Course evaluation 4.77 / 5.0 (95.4%).
For a full list of roles, publications, and skills, see the cv page.
Research as a Calling
Research is not separate from Mission2035. It is a scholarly calling entrusted to me in the present — a way to pursue truth with integrity, to serve patients and clinicians through responsible technology, and to prepare to teach and walk alongside students in higher education.
Published papers and manuscripts in preparation are listed on the publications page.