Research
Quantitative clinical research and AI methods across four surgical domains. Our methods standard is patient-level train/test splitting and discrimination metrics suited to imbalanced outcomes.
Burn care
Burn wound segmentation
Deep-supervision UNet++ that outlines burn wounds on clinical photographs, as a first step toward objective TBSA and depth assessment.
Inhalation injury prediction
Machine-learning prediction of inhalation injury from admission data (Burns, 2023), now extended to multimodal deep learning.
Burn outcomes from CGRD
Outcome research on the Chang Gung Research Database — transfusion strategy, infection, length of stay and reconstructive burden — with rule-based NLP for extracting injury details from clinical notes.
Rhinoplasty & craniofacial
Quantitative rhinoplasty
Landmark-based measurement of pre- and post-operative nasal form, with reliability-tested annotation tools, for reproducible aesthetic evaluation and outcome prediction.
Orthognathic bad-split prediction
CBCT and 3D-imaging models for predicting unfavorable fractures (bad splits) in sagittal split osteotomy, supporting orthognathic surgical planning.
Microsurgical reconstruction
Flap perfusion AI
Image-based assessment of free-flap circulation, from ICG fluorescence perfusion segmentation to a ResNet18 model that classifies flap status from a region of interest.