Catheter Navigation Vascular Digital Twin
Turn CT angiography data into a patient-specific vascular digital twin for catheter-navigation workflows.
The skill validates input paths, records logs, and states that patient data should not be committed; it nevertheless runs git clone, Dropbox download, extraction, symlink creation, and cache writes without explicit confirmation. It lacks checksums, permission isolation, sensitive medical-data handling, rollback, and cleanup guidance, so 14 points are deducted.
The ordered workflow, input checks, logging, artifact checks, and troubleshooting are useful. However, dependencies are fetched without a pinned revision, separate bash calls assume variable persistence, verification does not explicitly check the vessel mask and centerline, and the benchmark used only one attempt per task, so 12 points are deducted.
Triggers, input layout, cache outputs, and the non-fit synthetic-rendering case are reasonably clear. Local CT input avoids total dependence on downloads, but Chinese interaction, mainland-China network reachability, input variants, and clinical-use boundaries are not addressed, so 5 points are deducted.
The skill provides a name, version, Apache-2.0 license, owner, prerequisites, limitations, troubleshooting, and next-step guidance. The benchmark reports that version is placed at an ignored frontmatter level and that Instructions and Examples sections are missing; maintenance ownership, changelog, and update path remain incomplete, so 6 points are deducted.
The workflow plausibly completes CT preprocessing, vessel segmentation, and cache preparation, and the included evaluation reports correctness and activation results. Static review cannot verify actual medical outputs; the evaluation is small and segmentation still requires human review, so 9 points are deducted.
The pinned revision includes the skill, evaluation tasks, behavioral criteria, and one external-environment benchmark report. It lacks raw logs, reproducible artifacts, independent corroboration, and repository CI tests covering the key paths, so 6 points are deducted.
- Obtain explicit confirmation before downloading data, accessing the network, writing CACHE, or processing patient CT; keep real patient data out of repositories and logs.
- The Dropbox archive has no checksum and the external workflow repository is not pinned; production use should pin and verify dependencies.
- Checking only mu_volume.npy and metadata.json does not establish that the vessel mask and centerline are correct; require human or independent quality control.
- The segmentation output must not be used directly for clinical decisions.
What does this skill do, and when should you use it?
This skill supports the Isaac for Healthcare catheter-navigation workflow by preprocessing a CT NIfTI volume into an attenuation cache and segmenting the arterial tree into vessel-mask and centerline artifacts. It expects a subject directory containing ct.nii.gz and segmentations/, and documents the TotalSegmentator small public dataset as an option. The cache includes mu_volume.npy and metadata.json, with additional segmentation outputs after vessel processing. The resulting cache can feed the related viewport and DRR workflows.
Locates or downloads CT data, resolves the workflow repository, creates a timestamped run directory and logs, runs the i4h preprocess_ct command with HU saving, runs segment_vessels against the matching segmentation directory, and verifies that mu_volume.npy and metadata.json exist in the cache directory.
- An Isaac for Healthcare developer preparing patient-specific catheter-navigation inputs can use it to convert CTA data into a CT cache and vascular structures.
- A medical-imaging researcher with a TotalSegmentator subject directory can use it to create a reproducible vascular digital twin.
- A user preparing vessel masks and centerlines for the catheter-navigation viewport can run this as the upstream workflow.
- A user preparing attenuation data and vessel segmentation for DRR rendering can use it as the required preprocessing stage.
What are this skill's strengths and limitations?
- Covers the complete upstream path from CT preprocessing to vessel-mask and centerline generation.
- Provides reproducible shell commands, run-directory and logging conventions, and verification checks.
- Supports existing CT data and documents a download path for the TotalSegmentator small public dataset.
- Produces a cache intended for reuse by the related viewport and DRR workflows.
- Does not ship CT data; users must provide or download data and comply with its license.
- Requires the configured i4h-workflows environment and a matching segmentation directory.
- Large volumes may require at least 32 GB RAM, and segmentation may take several minutes.
- The source provides no platform test results or standalone test-suite information.
- The documented Zenodo mirror is throttled; the instructions prefer a Dropbox download URL.
How do you install this skill?
Use the repository README installation command:
npx skills add nvidia/skills --skill i4h-catheter-navigation-digital-twin --yes
The skill also requires i4h-catheter-navigation-setup to be completed and an available i4h-workflows checkout. The SKILL.md says the workflow may be cloned from https://github.com/isaac-for-healthcare/i4h-workflows when it is not found.
How do you use this skill?
Ask an agent with the skill loaded: “Preprocess this CT and segment the vessels to prepare ct_cache for the viewport.” Provide a subject directory containing ct.nii.gz and segmentations/, then set SUBJ and CACHE, for example:
SUBJ=/path/to/subject CACHE=/tmp/ct_cache
Run preprocess_ct followed by segment_vessels, then verify ${CACHE}/mu_volume.npy and ${CACHE}/metadata.json. The documented next steps are i4h-catheter-navigation-viewport or i4h-catheter-navigation-render-drr.