# Dicom Volume Quality V1

> Used to verify dicom-series-to-volume evidence packs by comparing JSON geometry evidence with the emitted NIfTI artifact. Not for clinical validation.

- Skill: `nvidia-medtech/dicom-volume-quality-v1` (Agent Skill, multi-file: 12 files)
- Install (CLI): `npx skillmds@latest add nvidia-medtech/dicom-volume-quality-v1`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nvidia-medtech/dicom-volume-quality-v1/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: Apache-2.0
- Author: NVIDIA-Medtech (https://skillmd.com/u/nvidia-medtech)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/nvidia-medtech/dicom-volume-quality-v1

---


# DICOM Volume Quality Verifier

## Purpose
- Used for deterministic second-pass review of a `dicom-series-to-volume` evidence pack.
- Checks that the source pack passed and that the declared NIfTI artifact matches reported shape, spacing, affine orientation, and voxel range evidence.
- Manifest I/O: inputs are `dicom_series_to_volume_evidence_pack`; outputs are `dicom_volume_quality_report`.

## Instructions
- Use this verifier only on an evidence pack directory produced by `skills/dicom-series-to-volume`.
- Run it through `eval_engine/run.py` when producing verifier evidence.
- The verifier entrypoint is `scripts/grade.py`; do not reimplement its checks in an agent prompt.
- Do not treat a pass as clinical validation of conversion quality.

## Available Scripts
| Script | Purpose | Arguments |
|---|---|---|
| `scripts/grade.py` | Primary verifier entrypoint declared by `skill_manifest.yaml`. | `EVIDENCE_PACK_DIR` |

## Prerequisites
- Runtime requirements: Python packages listed in `runtime.side_effects.pip_packages`.
- The input directory must contain `manifest.json`, `validation_summary.json`, `output.json`, and the NIfTI file declared by `output.path`.

## Limitations
- Audits the emitted NIfTI and JSON evidence; it does not re-read DICOM inputs.
- Does not prove clinical correctness of orientation or HU scaling.
- Does not repair non-canonical orientation, spacing, or voxel metadata.

## Troubleshooting
| Error | Cause | Fix |
|---|---|---|
| `target_skill_matches` fails | The fixture is not a `dicom-series-to-volume` evidence pack. | Re-run the verifier against the source skill pack directory. |
| `output_artifact_exists` fails | The source pack declares a missing or moved NIfTI path. | Keep the NIfTI inside the pack or use a repo-relative path that resolves from the repository root. |
| `shape_matches_nifti` fails | The JSON output no longer matches the saved NIfTI artifact. | Inspect the source pack and regenerate the evidence after fixing the converter. |

## Example

```bash
python verifiers/dicom_volume_quality_v1/scripts/grade.py \
  examples/evidence_packs/dicom_series_to_volume_pass

python eval_engine/run.py verifiers/dicom_volume_quality_v1 \
  --fixture examples/evidence_packs/dicom_series_to_volume_pass \
  --out runs/dicom_volume_quality
```