Medical Image Computing
We develop learning-based methods for segmentation, landmark detection, and localization across CT, CBCT, MRI, and X-ray images, with a particular focus on robustness when annotations are scarce — including semi-supervised, cross-domain, and prototype-based learning. Beyond methodology, we validate our systems in multi-center clinical studies with close clinical collaborators.
More in Publications.
Image Reconstruction
We study learning-based reconstruction for CBCT, DSA, and PET/CT — recovering high-quality 3D and 4D structures from sparse-view, low-dose, or artifact-corrupted acquisitions. Our recent work builds on implicit neural representations, radiative Gaussian splatting for dynamic vessel reconstruction, and 3D diffusion models for controllable, high-fidelity medical image generation.
More in Publications.
Digital Dentistry
We build full-stack AI for digital dentistry: tooth and alveolar bone segmentation from CBCT, clinical knowledge-informed tooth alignment for orthodontic planning, 3D teeth reconstruction from intra-oral photographs, and automated diagnosis of periodontal disease and dental caries. This line of work is carried out with close clinical collaborators and has produced several open benchmarks — CBCT segmentation, cephalometric landmarks (CephAdoAdu), tooth alignment, and caries detection (DVCT) — released on our Dataset page.
More in Publications · Open benchmarks in Dataset.