NVIDIA Launches Open-Source AI Model That Generates High-Resolution 3D Brain MRIs 33x Faster Than Previous Technology

NVIDIA launches NV-Generate-MR-Brain, an open-source AI model that synthesizes high-resolution 3D brain MRIs 33x faster than previous technology, trained on the world's largest open-source brain MRI dataset of 100,000 studies, aiming to break down data scarcity and privacy barriers in medical AI development.

• NVIDIA introduces NV-Generate-MR-Brain, a new open-source generative model built on the MAISI-v2 architecture that synthesizes high-resolution 3D brain MRI volumes with anatomical segmentation, supporting T1w, T2w, FLAIR, and SWI contrasts at resolutions up to 512×512×256.
• The model is trained on MR-RATE, the world's largest open-source multimodal brain MRI dataset, comprising 100,000 studies from over 83,000 patients and roughly 700,000 volumes paired with de-identified radiology reports and DICOM metadata, released under a CC-BY-NC license.
• The NV-Generate-CTMR framework addresses critical barriers in medical AI development—including data scarcity, privacy restrictions, and high annotation costs—by delivering 33x faster inference over prior diffusion models, flexible voxel and volume sizing, and fully open-source code, pretrained weights, and training configurations for immediate use on NVIDIA RTX GPUs.