Datasets:

---

nvidia

/

OpenH-RF

Like 54

Follow

NVIDIA 69.4k

Syncs the Resolve Stroke data cards, pipelines and figures on the Hub with the
reviewed state in the OpenH-RF GitHub repo (open-h/OpenH-RF PR #8).

Pipelines

• clinical/pipeline_PD_3d.yaml, phantom_flow/pipeline_PD_3d.yaml: grid raised
  from 160/64/64 to 200/80/80 (n_radial/n_azimuth/n_elevation, ~0.5 mm radial) and
  the tissue high-pass wall filter from 50 Hz to 100 Hz. The reconstruction scripts
  on GitHub load these configs directly, so the Hub copies need to match or the
  scripts silently reconstruct at the old settings.
• saddle/pipeline.yaml: comment only. zlims is on-axis depth; zea's
  polar_pixel_grid adds distance_to_apex to the radii itself, so the old note
  saying reconstruct.py offsets the apex no longer described the code.

clinical/ layout
The 20 clinical acquisitions share probe, sequence and file layout, so the data card
and pipelines move from clinical/SP02-Left-2/ up to clinical/, with an inventory
table covering all 20. clinical/SP02-Left-2/README.md, pipeline.yaml and
pipeline_PD_3d.yaml are removed in favour of the versions at clinical/.

Figures
Adds assets/ (52 files, ~4.9 MB) and links them from the cards: the main image
(3D power Doppler of clinical/SP03-Left at bolus +10 s, x-z MIP), B-modes,
reference maps, power-Doppler and MVI montages, the 21 saddle B-modes, and
per-acquisition thumbnails. The cards previously linked to PNGs that were never
uploaded, so those images were broken on the Hub.

Data cards
Version statements updated to zea 0.1.6; the computed reference maps described as
SYLVER outputs over the complete bolus provided as targets rather than clinical
ground truth. The top-level card gains the YAML frontmatter it was missing
(license, pretty_name, task_categories, tags) with size_categories: 100K<n<1M,
the combined frame total across the four sub-datasets (400,000 + 20,000 + 1,000 +
21 = 421,021), consistent with the per-sub-dataset cards.

Figures and pipelines are by the Resolve Stroke team (GitHub PR by @mxncr ); this
commit mirrors them to the Hub.

GitHub PR:

• tristan-deep/OpenH-RF#8 — Resolve Stroke: assets, main image, README and reconstruction updates

Update — common card layout. The cards now follow the layout shared by every OpenH-RF dataset: frontmatter with a name, a stand-alone title, the hero image with an italic caption linking the sample it shows, Description, Contributor(s), Creation Date and the standard CC BY 4.0 text, then a new Processing the Dataset section (the zea process command where the pipeline carries the whole reconstruction, and the reconstruct.py link) followed by Dataset Format, which opens with the zea version. Hard line breaks are gone so the Hub wraps paragraphs itself; the longer sections keep their content.

phantom_flow, phantom_mp and saddle now open with the figure each card already showed further down.

GitHub PRs:

• tristan-deep/OpenH-RF#28 — Restructure all data cards to the common README layout

Files changed: 5.

~ resolvestroke/README.md
~ resolvestroke/clinical/README.md
~ resolvestroke/phantom_flow/README.md
~ resolvestroke/phantom_mp/README.md
~ resolvestroke/saddle/README.md

Contributor sign off here: https://github.com/tristan-deep/OpenH-RF/pull/8