Abstract: |
| Reconstructing 3D CBCT (Cone-Beam Computed Tomography) volumes and bone shapes from ultrasparse X-ray projections can significantly reduce radiation doses while preserving informative 3D information. This seminar will introduce our recent advancements in this domain. First, we will present our DIFNet (Deep Intensity Field Network) approach, which aims to learn a deep intensity field mapping from 2D X-ray projections to 3D CBCT volumes. Next, we will discuss our C2RV (Cross-regional and Cross-view) method, a novel learning-based technique for sparse-view CBCT reconstruction. Finally, we will introduce our spatial-division augmented occupancy field method for bone shape reconstruction from biplanar X-rays. Through these three complementary approaches, we demonstrate significant progress in the field of 3D reconstruction from ultrasparse X-ray imaging, paving the way for reduced-dose medical imaging. |
Biography: |
| Xiaomeng Li is an assistant professor of Electronic and Computer Engineering at HKUST, where she also serves as the Associate Director of HKUST Centre for Medical Imaging and Analysis. Her primary focus lies in pushing the boundaries of AI to revolutionize the field of medicine. She is dedicated to developing innovative solutions that comprehend medical data, analyze intricate problems, and communicate at an advanced level. The goal is to craft AI doctors that can function autonomously or in collaboration with human physicians, thereby transforming healthcare delivery on a global scale and ensuring universal access to prompt, top-notch treatment. She has won multiple medical image analysis challenges and has received some best paper awards. Many of her innovations are widely used in tech and biotech industries. |