Turning clinical judgement into something measurable — and getting the number past the people whose judgement it replaces.

Since 2018 that has run through EEG, cephalograms, PET, whole-body CT, chest X-ray and cardiac CT, across segmentation, registration, reconstruction, landmark detection and GAN synthesis. Three became commercial products.

The model is rarely the hard part. On an automated tumour-response system, radiologists reading the same scans agreed less than 70% of the time — so most of the work was the evaluation, not the network. It shipped at 91% decision accuracy, clinically validated at Asan Medical Center.

• DivineTech - AI Team Lead
• Tesser - R&D AI Researcher
• MedicalIP - R&D AI Researcher
• Gachon University, MMMIL Lab - Graduate / Undergraduate Researcher

• medical-llm-eval - Evaluation harness for medical RAG: failure-mode case sets, stage-level error attribution, grader–human agreement
• medical-3d-viewer - nnUNetv2 segmentation, MPR and 3D rendering in a single browser view
• LungRegistration - VoxelMorph-based 3D lung registration; +90.2 mL lower-lobe volume change quantified by posture
• PET_sinogram - Physics-informed PET reconstruction, 257× fewer parameters than DeepPET
• cardiac-calcium-stent-extraction - Coronary calcification from cardiac CT with the metallic stent kept separate
• bone-fracture - X-ray fracture detection with Grad-CAM evidence

• Hong, Yong-Gi et al., Identification of Breathing Patterns through EEG Signal Analysis Using Machine Learning, Brain Sciences 11(3):293, 2021 - first author, SCI

• 3 patents filed - one as principal inventor
• 2 registered copyrights - commercialized

• 2025 SeSAC Hackathon - 2nd place, 2 of 32 teams, Seoul Mayor's Award

• Gachon University - M.Eng. Biomedical Engineering 🧠
• Gachon University - B.Eng. Biomedical Engineering 🧬

Python PyTorch nnU-Net MONAI SimpleITK VTK CUDA FastAPI Docker GitOps CI/CD

• Resume - Blog - LinkedIn - [email protected]