Senior research scientist at NVIDIA
MAISI: A Foundation Model for Accelerated, Anatomy-Aware High-Resolution 3D CT Synthesis
Medical imaging analysis is often limited by data scarcity, high annotation costs and privacy concerns. We introduce a unified framework that builds on our original Medical AI for Synthetic Imaging (MAISI) and its successor MAISI-v2 to tackle these challenges. MAISI combines a foundation volume compression network with a latent diffusion model conditioned via ControlNet on up to 127 anatomical structures to generate high-resolution 3D CT volumes (up to 512 × 512 × 768 voxels) with flexible dimensions and spacing. MAISI-v2 accelerates this process by integrating rectified flow for fast sampling achieving a 33× speed-up over baseline latent diffusion while adding a region-specific contrastive loss to strengthen alignment with segmentation conditions. In experiments across diverse body regions, our pipeline produces realistic, anatomically accurate volumes that improve downstream tasks, such as segmentation augmentation. We release code, training recipes, model weights and a GUI demo to support reproducibility and foster further development in the community.
Dr. Can Zhao is a Senior Applied Research Scientist at NVIDIA, where she leads the MAISI project — a foundation model for high-resolution medical image synthesis. Her research applies deep learning to medical imaging, with a focus on medical image synthesis and medical image quality enhancement. Dr. Zhao earned a bachelor’s degree in Semiconductor Physics from Tsinghua University in 2013. She then pivoted to medical image analysis, worked with Dr. Jerry Prince and received a Ph.D. in Electrical and Computer Engineering from Johns Hopkins University. Her thesis focused on medical image super-resolution and multi-modal image synthesis.
Since 2021, Dr. Zhao has organized the MICCAI Industrial Talk Series and served on the organizing committee of the MICCAI SASHIMI Workshop (2020–2023). In 2024, she chaired orals sessions at MICCAI and SPIE Medical Imaging. She is serving as the Industrial Liaison for the Women in MICCAI (WiM) board from 2025 to 2026.