DICOM CT Series to NIfTI Volume
Converts one CT DICOM series into a HU NIfTI volume with affine evidence.
The evidence shows Bash-only declaration, no network endpoints, caller-selected local output, and explicit engineering/non-clinical boundaries. However, there is no user confirmation, Bash remains broad, recursive reads expose Study/Series UIDs, dates, and descriptions that may be sensitive, and there is no de-identification, isolation, rollback, or cleanup mechanism. Eight points are deducted.
The documentation, manifest, script, schema, and verifier describe the main path, dependencies, JSON failures, and troubleshooting. Static code nevertheless does not truly reject multi-series or multi-frame input, may omit the NIfTI for inconsistent shapes, only reports spacing irregularities, and error responses may not satisfy the declared full output schema. No execution was performed, so the score remains below the static ceiling.
Inputs, outputs, single-series CT use, and major non-fit ranges are fairly clear, with positive and negative routing tasks. Trigger conditions, input preflight, clinical-data handling, and Chinese-language guidance are limited. Dependencies are local Python packages with no overseas service dependency, so no additional mainland-China reachability deduction applies.
SKILL.md, the manifest, schema, skill card, version 0.1.0, Apache-2.0 notices, and NVIDIA MedTech attribution provide useful structure. Missing are the recommended Examples section, a changelog, explicit maintenance ownership and update procedure; the supplied license metadata is NOASSERTION while the files state Apache-2.0, creating governance uncertainty. Points are deducted for these gaps.
The stated goal—converting one CT DICOM series to HU NIfTI while emitting geometry evidence—is concrete, and output fields plus the downstream verifier interface are specified. The supplied evaluation reports correctness and effectiveness on a small task set, but only two tasks are covered and static review did not execute or independently validate them; real-world DICOM coverage and comparative benefit remain limited. The score is kept within the static-review ceiling.
Fixed fixtures, a generator, schema, negative fixture, repeat-run configuration, paired verifier, and an evaluation report provide auditable structure. This review has only source-file evidence; the reported evaluation numbers and verifier results lack independently checkable CI artifacts here, and coverage is narrow. The score therefore stays below the static maximum.
- Do not use this converter for clinical care, diagnosis, regulatory submission, or production inference. Confirm a single CT series and inspect the JSON, affine, axcodes, shape, spacing, and HU range.
- Real DICOM may contain PHI; the script emits UIDs, dates, and descriptions. Apply approved de-identification, access controls, and evidence-directory protection before execution.
- Preflight multi-frame, compressed-transfer, mixed-series, inconsistent-shape, and irregular-spacing inputs separately; do not rely solely on direct script execution.
What does this skill do, and when should you use it?
This skill converts one CT DICOM series folder into a Hounsfield-unit NIfTI volume. It sorts slices, applies DICOM rescale parameters, and derives an affine from orientation and spacing metadata. The run produces a compressed NIfTI file and a JSON summary, which should be checked with the paired dicom_volume_quality_v1 verifier. It is limited to single-series, non-multiframe, uncompressed CT data and is not intended for clinical deployment.
Reads one DICOM series directory; sorts slices by ImagePositionPatient; applies RescaleSlope and RescaleIntercept; builds an affine in NIfTI/RAS coordinates from orientation and spacing tags; runs scripts/series_to_volume.py; and writes a .nii.gz volume plus JSON summary fields including slice count, series instance UID, shape, spacing, axis codes, affine, HU range, output path, and conversion time.
- A medical-imaging engineer needs to convert one CT series into HU-scaled NIfTI data for downstream processing.
- A data scientist needs slice ordering, voxel spacing, and affine evidence alongside the converted volume.
- A medical-AI developer needs to validate orientation with a downstream gate such as expected_axcodes before segmentation.
- An evaluator needs a trusted conversion run with preserved JSON output and verifier evidence.
What are this skill's strengths and limitations?
- Provides a concrete CT-to-HU-NIfTI conversion workflow.
- Emits inspectable affine, axis-code, spacing, and HU-range evidence.
- Documents important input and deployment boundaries.
- Supports Bash execution and a trusted-run path.
- Accepts only one CT series; multi-series input is rejected at preflight.
- Does not support multiframe DICOM files with NumberOfFrames greater than 1.
- Does not support JPEG, JPEG2000, or RLE compressed transfer syntaxes.
- Does not auto-reorient voxels and is not for clinical deployment, autonomous diagnosis, regulatory submission, or production inference.
- The supplied material does not specify Python package versions or an independent test suite.
How do you install this skill?
Install the specific skill through the skills CLI flow documented in the repository README:
npx skills add nvidia/skills --skill dicom-series-to-volume --yes
The CLI prompts for the skill and installation destination. The supplied material does not document a standalone package or fixed installation directory.
How do you use this skill?
Run from the repository root and keep outputs under a caller-provided run directory. Read skill_manifest.yaml before changing arguments or validation, then run:
python scripts/series_to_volume.py PATH_TO_DICOM_DIR --output PATH_TO_OUT.nii.gz
For a trusted run, use:
python -m eval_engine.run_trusted skills/dicom-series-to-volume --fixture PATH_TO_DICOM_DIR --out runs/dicom_series_to_volume_trusted
Inspect the emitted JSON and run the paired dicom_volume_quality_v1 verifier before treating the result as evidence.
How does this skill compare with similar options?
The SKILL.md names dcm2niix as the vetted converter to use for clinical deployment. This skill therefore fits constrained single-series engineering conversion and evidence checking, not clinical conversion workflows.