skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

nv-generate-ct-rflow

Generate synthetic CT volumes and segmentation masks with the NV-Generate-CTMR rflow-ct pipeline.

3.2k|370|Updated Feb 25, 2026
Search Tags:#medical imaging#data augmentation#NIfTI#diffusion model#nv-generate-ctmr#segmentation mask#synthetic ct
One-click install
npx skills add https://github.com/NVIDIA/skills --skill nv-generate-ct-rflow
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: nv-generate-ct-rflow
Source: https://github.com/NVIDIA/skills/tree/main/skills/nv-generate-ct-rflow
Command: npx skills add https://github.com/NVIDIA/skills --skill nv-generate-ct-rflow

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires nibabel, numpy, typer, torch, monai, matplotlib, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Medical imaging researchers need paired synthetic CT volumes and segmentation masks for research, data augmentation studies, and pipeline development, but running the NV-Generate-CTMR rectified-flow synthesis pipeline correctly requires careful config staging, anatomy validation, and output verification.

Core Features & Use Cases

  • Paired CT and mask synthesis: Runs the upstream rflow-ct inference entrypoint with a JSON config override, producing NIfTI image/label pairs plus an auditable result JSON.
  • Preflight validation and cost gating: Checks anatomy names against the 132-class label dictionary, body region, FOV constraints, CUDA availability, and estimated VRAM/wall-time before launching expensive GPU inference.
  • Visual summary card: Writes a summary.html with mid-slice axial/coronal/sagittal triptychs and label overlays for quick inspection without 3D Slicer.
  • Use Case: A researcher asks the agent to generate a synthetic abdomen CT with a controllable hepatic tumor from a config file; the skill validates the request, runs the pinned upstream pipeline on a CUDA GPU, and returns verified image/mask pairs with label mapping evidence.

Quick Start

Ask your agent to generate a synthetic CT volume and paired mask from your config file using the nv-generate-ct-rflow skill with a chosen output directory and random seed.

Frequently Asked Questions about nv-generate-ct-rflow

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate synthetic CT volumes with segmentation masks?▼

Run scripts/run_rflow_ct.py with a JSON config override as the first argument, plus --output-dir and --random-seed. The wrapper stages the config into the NV-Generate-CTMR checkout and invokes the upstream rflow-ct inference entrypoint, producing NIfTI image/label pairs.

How do I validate a CT generation request before running inference?▼

Pass --preflight-only to run_rflow_ct.py. Preflight checks config schema bounds, anatomy names against the upstream label dictionary, body region, FOV minimums, dataset presence, CUDA availability, and estimated VRAM and wall-time without launching inference.

What GPU do I need for rflow-ct CT synthesis?▼

rflow-ct requires CUDA with roughly 16 GB VRAM minimum for the default 256 cubed output size. Larger outputs such as 512x512x768 need an A100 or H100, and there is no CPU fallback in the upstream code path.

Can synthetic CT volumes be used as production training data?▼

No. Output volumes are synthetic and are not safe as training data for production medtech models without an independent quality review. They are also not for clinical deployment, clinical interpretation, autonomous diagnosis, or regulatory submission.

Why does controllable tumor generation produce missing or tiny tumor labels?▼

Small controllable_anatomy_size requests such as 0.2 can produce absent or extremely small label-23 lung tumor components for some seeds. Prefer a controllable size around 0.5 or larger, and check output.output_label_mapping in the result JSON.

Related Skills

Explore More Education & Research Skills→
sayalinvidiasayalinvidia

nv-segment-ctmr

Run NV-Segment-CTMR MONAI bundle segmentation and output a labeled map with JSON report.

Community
Intermediate
Quill-AgentQuill-Agent
3

segment-anything-model

Segment objects in images using points, boxes, or masks.

Community
Intermediate
KoStardKoStard
917

forgecad-model-grader

Grade ForgeCAD models against requirements with evidence-based 0-10 scoring.

Community
Advanced
NousResearchNousResearch
239.1k

stable-diffusion

Generate images from text prompts using HuggingFace Diffusers pipelines.

Official
Advanced