Research
A unified framework linking what we plan before surgery, what we do during surgery, and what happens to the patient afterward. Built around robotic total mesorectal excision (TME) and left-sided colorectal resection.
1. Planning
Anatomical risk stratification from preoperative imaging. Automated CT-based pelvimetry quantifies the mid-pelvic surgical workspace before the patient enters the operating room.
- A fully automated CT-based pelvimetry pipeline for quantifying mid-pelvic surgical workspace in rectal cancer. Int J CARS. 2026. doi
Open-source: ctpelvimetry · GitHub · Demo
2. Proficiency
Quantitative assessment of robotic technical skill. Video-based risk-adjusted CUSUM and machine learning models replace conventional case-count thresholds for learning curve evaluation.
- Technical proficiency assessment of robotic intracorporeal single-stapling colorectal anastomosis using video-based RA-CUSUM. Int J Colorectal Dis. 2026. doi
- Machine learning–based risk modeling for safety-focused learning curve assessment in robotic left-sided colorectal cancer surgery. J Robotic Surg. 2026. doi
- Robotic intracorporeal single-stapling versus double-stapling anastomosis in left-sided colorectal cancer: a propensity score-weighted pilot study. J Robotic Surg. 2025. doi
3. Technique & Outcomes
Novel intracorporeal anastomotic techniques and their oncologic, functional, and short-term outcomes. The Robotic intracorporeal Single-Stapled Anastomosis (RiSSA) technique reduces stapler firings and bowel manipulation.
- The robotic intracorporeal single-stapled anastomosis (RiSSA) technique in robotic left-sided colorectal resection: a technical note. Ann Coloproctol. 2025. doi
- Robotic intracorporeal single‐stapled anastomosis (RiSSA) and natural orifice specimen extraction (NOSE) in total mesorectal excision for rectal cancer—A video vignette. Colorectal Disease. 2024. doi
- Redo Robotic Low Anterior Resection by Natural Orifice Intracorporeal Anastomosis With Extraction. Diseases of the Colon & Rectum. 2025. doi
Tooling
All quantitative work is reproducible. Pipelines run in Python (PyTorch, scikit-learn) on GCP (NVIDIA L4 GPU). Source available on GitHub.