Master's Student
Seoul National University
M.E in Interdisciplinary Program of Bioengineering
June 2020 - September 2022
Seoul, Korea
Research Focus
Established foundational expertise in medical image analysis and AI systems through pioneering research in AI-Enhanced Augmented Reality Navigation Systems. This work in ENT (Ear, Nose, Throat) applications and dental imaging provided crucial technical foundations that directly inform current Vision-Language model research at UCLA.
Master's Thesis: "Development of an AI-Enhanced AR Navigation System for Precise Mastoidectomy" - A comprehensive system that established core competencies in computer vision, deep learning, and medical imaging that continue to drive current research directions.
Legacy Impact: While no longer actively working in surgical navigation, the deep learning segmentation, medical imaging processing, and precision AI system development skills from this period form the technical foundation for current Vision-Language Foundation model work.
Research Projects
ENT Navigation System for Mastoidectomy
Clinical Challenge: Chronic otitis media (COM), mastoiditis, and cholesteatoma are among the most common middle ear inflammatory conditions requiring precise surgical intervention. Traditional surgical techniques have a steep learning curve, and navigation technology is essential for immediate anatomical verification.
Technical Innovation
- AR-Microscope Integration: Seamlessly integrated augmented reality with surgical microscopes for real-time anatomical visualization
- High-Precision Matching: Developed specialized ear matching algorithms achieving sub-millimeter accuracy requirements
- Deep Learning Registration: Implemented AI-based registration between optical trackers and depth cameras
- Real-time Processing: Optimized algorithms for real-time performance during surgical procedures
Clinical Significance
The AR-microscope integration provides surgeons with enhanced visualization capabilities for complex cases involving inherent abnormalities or distorted anatomical structures. This technology significantly reduces surgical risk and improves precision in delicate ear surgeries.
Research Outcomes
- 2 Patents granted for navigation apparatus and registration methods
- 6+ Conference papers published in top medical conferences
- Clinical validation through collaboration with ENT surgeons
- Technology transfer potential for commercial surgical systems
Unintended Motion Detection System
Safety Challenge: Unintended motion by surgeons during delicate ear procedures can cause injury to critical anatomical structures such as the tympanic membrane and ossicular chain.
AI-Powered Safety Solution
- Real-time Tracking: Continuous monitoring of surgical tool positions and movements
- Predictive Analysis: AI algorithms predict tool trajectory and identify potential collision risks
- AR Visualization: Display current tool location and safe/expected positions in next frames
- Collision Avoidance: Automated warnings and guidance to prevent medical mistakes
Technical Achievement
Successfully integrated computer vision with surgical workflow to create a comprehensive safety system that enhances surgical precision while maintaining natural workflow for surgeons.
OCT Image Segmentation - Alveolar Bone Level Assessment
Clinical Need: Accurate assessment of alveolar bone levels is essential for periodontal diagnosis. Traditional CBCT imaging involves high radiation doses, particularly concerning for pediatric patients.
OCT-Based Solution
- Non-invasive Imaging: Leveraged Optical Coherence Tomography as a safer alternative to CBCT
- Deep Learning Segmentation: Developed automated algorithms for precise alveolar bone crest identification
- Real-time Processing: Enabled immediate diagnostic feedback during clinical examinations
- Quantitative Analysis: Automated measurement of bone levels with high accuracy
Research Impact
Co-first author publication in Biomedical Optics Express, demonstrating significant advancement in non-invasive periodontal diagnosis techniques.
Superpixel-based Graph Convolutional Network
Technical Challenge: Traditional encoder-decoder CNNs struggle with precise boundary preservation in semantic segmentation due to information loss during downsampling and upsampling operations.
Graph-Based Innovation
- Pooling-Free Architecture: Eliminated pooling layers to preserve object shape information
- Superpixel Preprocessing: Segmented images into meaningful clusters based on RGB values
- Graph Representation: Treated superpixels as graph nodes for GCN processing
- Dual Convolution Strategy: Combined spectral and spatial convolutions for comprehensive feature capture
- Novel Loss Function: Introduced Superpixel Penalty Loss addressing class and size imbalances
Evaluation
Tested on UAVid dataset with ambiguous object boundaries. While not achieving state-of-the-art performance, demonstrated comparable pixel classification ability and successfully expanded GCN concepts into semantic segmentation domain.
OCT Speckle Noise Reduction
Image Quality Challenge: Speckle noise in OCT imaging is inherent multiplicative noise caused by light wave scattering, reducing contrast and limiting effective resolution.
Advanced Denoising Approach
- Deep Learning Framework: Replaced traditional B-scan averaging with AI-based denoising
- Artifact Prevention: Eliminated registration errors and motion artifacts common in averaging methods
- Quality Enhancement: Improved both axial and lateral image resolution
- Real-time Processing: Maintained clinical workflow efficiency
OCT Gingival Sulcus Segmentation
Clinical Application: Gingival sulcus depth measurement is crucial for periodontal diagnosis. Traditional manual probing causes patient discomfort and lacks precision.
OCT-Based Measurement System
- Non-invasive Approach: Eliminated patient discomfort from manual probing
- High-Resolution Imaging: Achieved superior resolution compared to CT, MRI, and ultrasonic imaging
- Real-time Capability: Enabled immediate diagnostic feedback
- Automated Analysis: Precise sulcus depth measurement through AI segmentation
Research Publications from SNU
Journal Publications
Co-First Author - Biomedical Optics Express
Sul-Hee Kim*, Jin Kim*, Su Yang, Sung-Hye Oh, Seung-Pyo Lee, Hoon Joo Yang, Tae-Il Kim, and Won-Jin Yi
"Automatic and Quantitative Measurement of Alveolar Bone Level in OCT Images Using Deep Learning"
Conference Publications
EMBC 2022 - First Author
"A Faster Deep learning-based Registration for Ear Surgery"
Jin Kim, Da-El Kim, Min-Hyuk Choi, Su Yang, Se-Ryong Kang, So-Young Chun, Won-Jin Yi
EMBC 2021 - First Author
"A Deep Learning-based Method for Tooth Segmentation on Panoramic Dental X-ray Images"
Jin Kim, Su Yang, Min-Hyuk Choi, Sang-Jeong Lee, Bo-Soung Jeoun, Geonsoo Kim, Won-Jin Yi
CICS 2021 - First Author
"SANet: Self-Attention U-Net for Binary Tooth Segmentation"
Jin Kim, Su Yang, Min-Hyuk Choi, Bo-Soung Jeoun, Won-Jin Yi
ICS 2022 - First Author
"A Registration Method between Optical Tracker and Depth Camera Using Deep Learning"
Jin Kim, Da-El Kim, Min-Hyuk Choi, Su Yang, Se-Ryong Kang, So-Young Chun, Jun-Young Yoo, Seung Noh, Won-Jin Yi
Collaborative Publications
"CAPPU-Net: A Convolutional Attention Network with Pyramid Pooling for Segmentation of Middle and Inner Ear Structures in CT Images"
Geonsoo Kim, Bo-Soung Jeoun, Su Yang, Jin Kim, Sang-Jeong Lee, Won-Jin Yi - EMBC 2021
"SinusNet: Label-Free Segmentation of Maxillary Sinus Lesion in CBCT Images"
Da-El Kim, Su Yang, Se-Ryong Kang, Jin Kim, So-Young Chun, Min-Hyuk Choi, Won-Jin Yi - MIDL 2022
Patents & Intellectual Property
AR Navigation Apparatus
"NAVIGATION APPARATUS AND METHOD BASED ON AUGMENTED REALITY MICROSCOPE"
Seon Tae Kim, Dong Kyu Kim, Wonjin Yi, Jin Kim, Bo Soung Jeoun
Augmented reality-based navigation system for enhanced surgical precision using microscope integration. Core technology from Master's thesis research.
Navigation System with Image Registration
"NAVIGATION SYSTEM USING SPRINT AND IMAGE REGISTRATION METHOD USING THE SAME"
Seon Tae Kim, Dong Kyu Kim, Wonjin Yi, Jin Kim, Bo Soung Jeoun
Advanced navigation system utilizing sprint technology and sophisticated image registration methods for surgical applications.
Key Achievements at SNU
Master's Degree
M.E. in Interdisciplinary Program of Bioengineering with Merit-based Scholarship (100% tuition waiver)
2 Patents Granted
Secured intellectual property for AR navigation systems and image registration methods
6+ Publications
First author on multiple conference papers and co-first author on journal publication
Clinical Translation
Research directly applicable to surgical procedures with collaboration from medical professionals