[Histopathology P0] Smoke test QuPath selected ROI -> registered LabConstrictor tool -> RGB-to-scalar -> FiberFeatures results #49

Goal

Deliver the first executable, end-to-end QuPath-to-MorphoBricks pilot on one small ROI. Reuse the existing generic LabConstrictor worker/manifest bridge; do not build a QuPath-specific feature extractor or wait for WSI streaming.

QuPath 0.7 selected annotation / viewport -> current bridge TIFF + optional RegionOf mask -> registered Python tool -> morphobricks-normalization RGB-to-YX scalar -> existing FiberFeatures API -> table + image outputs -> QuPath results window.

This is a small-ROI smoke test only. Attaching measurements to existing QuPath objects and large-slide file-based results binding are later bridge work; the pilot must still leave a useful reproducible results bundle.

Code-grounded starting point

• LabConstrictor–QuPath README: registered apps/tools via JSON-lines worker, selected-annotation/current-viewport image areas, TIFF export, results table/images/points/shapes. QuPath supports the folder and file input types too.
• Groovy bridge implementation: LcArea and prepareImageInput freeze/crop the selected plane/area and refuse inputs over 100M px / 4 GiB, addResult displays feature tables and image previews.
• Selected annotation(s) currently exports a BOUNDING RECTANGLE, not the true polygon unless the tool also requests a RegionOf annotation mask. Do not infer from bounding-box pixels that all are collagen/ROI interior.
• LabConstrictor-Tools protocol supports image, labels, table and file outputs; use its existing declarations and hosts: [qupath] registration as appropriate.
• Conversion implementation owner: morphobricks-normalization #94, not the bridge or FiberFeatures.
• FiberFeatures pilot & validation: fiberfeatures #210; first outputs from unchanged run_fiberfeatures_pipeline.

Minimal steps (do in this order)

1. Register one small LabConstrictor-compatible Python demonstration tool (app ownership/packaging to agree) with declared RGB image input, optionally the exact selected-ROI mask, explicit stain profile/method choice, and calibrated pixel-size parameters when available. It should be discoverable under QuPath without any new hard-coded app name in the bridge.
2. Open a known 8-bit sRGB Masson's blue-trichrome crop/slide, select exactly one annotated ROI, run using Image area: Selected annotation(s). Confirm source file/series, z/t plane, region origin, actual RGB channel order/axes, real colour data (not QuPath display LUT), saved crop dimensions and physical calibration. An ambiguous RGB image type or unknown required calibration must be refused, not guessed.
3. If the ROI is nonrectangular, pass its actual mask via the existing region-mask mechanism and validate mask/image alignment. Do not pretend the crop bounding box equals the ROI.
4. Call the normalization-owned rgb_to_scalar options (initially luminance, signed CIELAB blue, bpx methyl-blue stain component), then unchanged FiberFeatures 2D core with identical physical scales/configuration and a common ROI.
5. Return a small table of measurements (including method, ROI/source identity, applicability/NaN reasons) and image outputs for scalar map, ridge support and orientation/coherence. Confirm the bridge opens table and previews, and keeps a run manifest with exact revisions, geometry, source colour/profile, calibration and conversion method.
6. Run a noncollagen / nucleus-rich control on the same slide. Conclude pass/fail for correct plumbing and structural specificity, not clinical/SHG validation.

What to verify adversarially

• QuPath RGB serialization: the 3 RGB samples must remain one RGB imaging unit, not three independent fluorescence channels, nor a (Y,X,3) pseudo-3D volume; confirm the actual exported TIFF array and declared axes (QuPath file import and Python reader may disagree on channel ordering).
• True, consistent ROI support in all three conversion comparisons; RegionOf must be on the same z/t as the exported RGB.
• Known dy,dx (or explicitly pixel-only analysis); do not use the QuPath bridge's single square-pixel field to silently replace non-square calibration.
• Existing FiberFeatures preprocess_image internally percentiles each scalar image: do not interpret derived output as absolute stain concentration.
• Freeze ROI and target image at run start; changing slide/selection during a delayed worker run cannot rebind the outputs (existing fixes for QuPath #16, #21).
• No unnecessary full-slide export and no modifications to arbitrary installed image files or QuPath annotations for this P0.

Acceptance

• A developer can run a documented QuPath 0.7 GUI fixture (small RGB slide + one ROI), producing a real feature table and preview, through an installed/registered tool with no manual Python subprocess invocation.
• Recorded TIFF/mask pixel and ROI coordinate tests, RGB channel/axes assertions, and source-calibration provenance. Failed inputs give explicit refusal messages.
• Runs without CellPatcher's still-planned WSI lazy readers, without changing the FiberFeatures measurement implementation, and without assuming new QuPath measurement-attachment support.
• Independent end-to-end smoke test executed (distinguish CI automation from human GUI testing). Include OS/QuPath version and whether verified on that platform.

Not in scope: WSI/Ome-Zarr lazy tile orchestration, automatic attachment to existing ROIs, large spatial heatmap overlays, colour tensor/new ridge features.