ClinGeom: execute pinned highdicom #454 frame-identity witness

[lindicaphxag-tech]

Goal

Turn the source-audited highdicom #454 FrameOfReferenceUID gap into a public, pinned, executable witness.

Frozen runtime

• highdicom PR #454 head: 4524bb0b0c8b3141f6f5b26d047b8da43c69e61e
• MONAI: 1.6.1
• Python: 3.12

Expected witness

The same run must show:

1. Volume -> MetaTensor -> Volume affine geometry round-trips;
2. automatic FrameOfReferenceUID does not round-trip;
3. two source volumes with distinct UIDs can produce equal MetaTensor affines;
4. explicitly supplying frame_of_reference_uid restores the identity.

The workflow uploads the JSON witness as an Actions artifact.

Claim boundary

This is a public execution witness against a pinned open PR head. It is not maintainer confirmation, upstream adoption, or a claim that highdicom has accepted this diagnosis.

Executed result

Public Actions run: https://github.com/lindicaphxag-tech/kaggle/actions/runs/37475115875

Pinned execution result:

{
  "geometry_roundtrip_max_abs_error": 0.0,
  "automatic_roundtrip_uid_a": null,
  "automatic_roundtrip_uid_b": null,
  "automatic_uid_preserved": false,
  "explicit_uid_preserved": true,
  "same_meta_affine_for_distinct_source_uids": true
}

The uploaded artifact digest is:

sha256:7d5ff2768428049acf40f1e70292cbfb83de9cbbcb420b34b070ad91fd1297f4

Interpretation: numerical voxel-to-world geometry round-trips exactly on this witness, while DICOM physical-frame identity is not automatically carried through the current pinned Volume ↔ MetaTensor bridge. Explicit caller-supplied identity still works.

This remains a public execution witness, not upstream maintainer confirmation.

Persistent review artifacts

• Executed result record: research/medical_geometry/upstream_repros/HIGHDICOM_454_RESULT.md
• Minimal patch proposal: research/medical_geometry/upstream_repros/HIGHDICOM_454_PATCH_PROPOSAL.md
• Executable witness: research/medical_geometry/upstream_repros/highdicom_454_frame_identity.py

These are intentionally kept independent of the ClinGeom method claims so an upstream reviewer can evaluate the interoperability issue without adopting ClinGeom.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}