# NVIDIA MedTech Open Models

> Open-source foundation models for medical imaging and physical AI — segmentation,
> synthetic CT/MR generation, chest-X-ray reasoning, raw-signal reconstruction, and
> surgical world models. Many are runnable by a coding agent (Claude Code, Cursor,
> Codex) via the open Agent Skills spec: install a skill and the agent runs the model
> correctly first try. Engineering and research use only — NOT a medical device, not
> for diagnosis or clinical decision-making.

## Run with an agent
- [Install a skill](https://github.com/NVIDIA-Medtech/medical-AI-skills): `npx skills add NVIDIA-Medtech/medical-AI-skills`
- [Agent Skills spec](https://agentskills.io): How agent skills work and which agents support them
- Proof: agents with the skill produced a valid first-shot command 79/81 times vs 0/81 with the README alone

## Models with agent skills
- [NV-Segment-CT](https://nvidia-medtech.github.io/models/nv-segment-ct.md): 132-class CT segmentation (VISTA3D) — run + finetune skills
- [NV-Segment-CTMR](https://nvidia-medtech.github.io/models/nv-segment-ctmr.md): 345+ class CT & MRI segmentation
- [NV-Generate-CT](https://nvidia-medtech.github.io/models/nv-generate-ct.md): 3D synthetic CT generation (+ paired masks)
- [NV-Generate-MR](https://nvidia-medtech.github.io/models/nv-generate-mr.md): 3D synthetic MRI generation (rectified flow)
- [NV-Generate-MR-Brain](https://nvidia-medtech.github.io/models/nv-generate-mr-brain.md): synthetic brain MRI — run + finetune skills
- [NV-Reason-CXR](https://nvidia-medtech.github.io/models/nv-reason-cxr.md): chest-X-ray clinical reasoning VLM

## DICOM utility skills (no GPU)
- [DICOM Metadata Extract](https://nvidia-medtech.github.io/skills/dicom-metadata-extract.md): extract header metadata + flag PHI tags from one DICOM file
- [DICOM Series Preflight](https://nvidia-medtech.github.io/skills/dicom-series-preflight.md): header-only checks on a series folder before conversion or inference
- [DICOM Series to Volume](https://nvidia-medtech.github.io/skills/dicom-series-to-volume.md): convert a single CT DICOM series to a HU NIfTI volume

## Other models
- [GR00T-H](https://nvidia-medtech.github.io/models/gr00t-h.md): healthcare physical-AI vision-language-action model
- [Cosmos-H-Surgical](https://nvidia-medtech.github.io/models/cosmos-h-surgical.md): surgical world model
- [Cosmos-H-Surgical-Simulator](https://nvidia-medtech.github.io/models/cosmos-h-surgical-simulator.md): action-conditioned surgical simulation
- [NV-Raw2Insights-MRI](https://nvidia-medtech.github.io/models/nv-raw2insights-mri.md): raw MRI signal reconstruction
- [NV-Raw2Insights-US](https://nvidia-medtech.github.io/models/nv-raw2insights-us.md): raw ultrasound processing

## More
- [Home](https://nvidia-medtech.github.io/index.md): All models + the Agent Skills section
- [News](https://nvidia-medtech.github.io/news.md): Releases and announcements
- [Full corpus](https://nvidia-medtech.github.io/llms-full.txt): Key pages concatenated as Markdown