The model operates in the compressed latent space of a frozen mask autoencoder, generating 256^3 volumetric label maps.
3D diffusion model generates body CT masks with controllable anatomy, tumors, demographics
NV-Generate-BodyMask uses a 19-dimensional conditioning vector to steer organ sizes, lesion burden, and patient attributes in synthetic whole-body segmentation masks.
Big Tech
Can Zhao · Pengfei Guo · Dong Yang · Yufan He · Yucheng Tang · Ahmed Harouni · +5 more
NVIDIA · National Institutes of Health · University of Arkansas for Medical Sciences
Research Digest··2 min read
Thread:Diffusion Avatar Generation
The authors present NV-Generate-BodyMask, a 3D latent diffusion model that synthesizes whole-body CT segmentation masks conditioned on explicit organ size quantiles, five tumor types, and four demographic variables.
§