Synthesize Realistic 3D Medical Images at Scale to Ship Pre‑Trained Models

NVIDIA has released NV-Generate-CTMR and NV-Generate-MR-Brain, open-source frameworks for synthesizing realistic 3D CT and MRI volumes at scale. Built on the MAISI architecture family, MAISI-v2 uses Latent Rectified Flow for 33x faster inference compared to DDPM-based approaches. The framework supports flexible voxel sizes, variable volume dimensions, and whole-body coverage in a single model. A companion dataset, MR-RATE, containing 100,000 brain MRI studies from 83,000+ patients (~700,000 volumes) with paired radiology reports, has been released under CC-BY-NC for research. Models are available with commercial-friendly open source licenses, runnable on NVIDIA RTX GPUs, and include pretrained weights and inference scripts for immediate use in data augmentation, privacy-preserving data sharing, and downstream medical AI training pipelines.