NVIDIA just open-sourced a model that generates synthetic brain MRIs.

NV-Generate-MR-Brain:

→ 3D latent diffusion model
→ Generates T1, FLAIR, T2, and SWI scans
→ 512x512x256 resolution at 0.45mm precision
→ 33x faster inference than previous gen
→ Trained on 40,000 MRI scans
→ Commercial use ready

Why this is huge:

Medical AI is bottlenecked by data. Patient privacy, rare conditions, expensive annotations.

This model generates unlimited synthetic brain scans for training. No real patients. No privacy issues.

Coming soon: cross-modality synthesis. Generate a FLAIR scan from a T1. Patients might only need one scan instead of four.

Released today at GTC 2026