Skip to content

Check tile quality

Quality control keeps background, blurred areas and weakly stained regions out of the prediction. Review the retained tissue before moving on to molecular exploration.

Prepare high-quality tissue inputs

The whole section The whole section in context, stained tissue meeting the pale mount and bare glass around it, with three cyan region markers.
Exclude glass and background The left end of the section, where the stained tissue gives way to pale mount above it and bare glass below: the background that should not become a model tile.
Keep representative tissue Representative lymphoid tissue at full resolution, densely stained with visible nuclei and little empty background.
Inspect one model-sized tile One square model-sized tile cropped from the centre of the representative lymphoid tissue field.
An open-access slide of a melanoma metastasis in a lymph node. The last panel is one tile at the size the model reads.

Read the panels in order: the whole section, the glass to exclude, representative tissue, and one tile at full detail.

Start with a small representative region: pack it with pack_tiles, read rejected_by_reason, and inspect examples of each reason before processing the whole slide.

The one value to get right

Read mpp from the slide's metadata, and do not assume it. A tile declared at the wrong scale is judged against the wrong physical size, so the verdict is on the wrong thing. Install and prepare a slide is where you type it.

What the verdict tells you

The verdict keeps glass, blur and faint stain out of your result, and names what it set aside.

Quality check Why it matters What to inspect
Tissue coverage Excludes mostly empty regions. Tissue boundaries and background.
Image detail Identifies regions without enough usable detail. Focus at full resolution.
Staining Identifies regions with weak H&E signal. Faint tissue and scanner colour.

The report gives retained and excluded counts, with reasons for exclusion. The package reference documents the returned fields.

More tiles than one submission takes

Choose a region relevant to your question if the slide exceeds a submission limit. Keep a contiguous area so neighbouring tissue remains available for spatial interpretation.

The spatial selection helper can select a dense region before upload. Aurora predicts on every tile you submit.

Next

  • Ask for a key

    Request programmatic access, sign in and check your setup. Get access

  • From slide to result

    Follow one H&E slide to its spatial gene expression result. From slide to result

  • Your slides stay with you

    Run DeepSpot-H on your own machine and send the numbers it produces. Your slides stay with you