Pathology inputs
Slide URL, workflow, tile size, magnification, quality filters, and stain normalization.
You provide a whole slide image and pick a workflow. The two inputs that most shape the result are the slide itself (its format and quality) and the workflow you choose — everything else tunes tiling, quality filtering, and normalization. The slide is read from workspace storage as an S3 URI, so upload it there first. See the quickstart for the end-to-end flow.
Required input
| Parameter | Type | What it does |
|---|---|---|
slide_url | string (S3 URI) | Path to the whole slide image in workspace storage. Supported: SVS, TIFF, NDPI, or standard images (PNG/JPG). Required. |
Workflow
| Parameter | Type | Default | What it does |
|---|---|---|---|
workflow | enum | tile_extraction | Which analysis to run: tile_extraction, tissue_detection, cell_segmentation, classification, or feature_extraction. |
Only tile_extraction, tissue_detection, and feature_extraction run. The schema accepts classification and cell_segmentation, but the backend rejects both with an "Unsupported workflow" error.
Tile extraction parameters
| Parameter | Type | Default | Range | What it does |
|---|---|---|---|---|
tile_size | integer | 256 | 64–1024 | Edge length in pixels of each square tile. |
overlap | integer | 0 | 0–128 | Pixel overlap between adjacent tiles. |
magnification | enum | 20x | 5x / 10x / 20x / 40x | Magnification level to tile at. |
max_tiles | integer | 1000 | 1–10000 | Cap on how many tiles to keep. |
Quality filtering
Tiles that fail either filter are discarded, so a tile set can be much smaller than the raw grid.
| Parameter | Type | Default | What it does |
|---|---|---|---|
tissue_threshold | float | 0.5 | Minimum fraction of a tile that must be tissue (non-background) to keep it. |
blur_threshold | float | 100.0 | Minimum Laplacian variance to keep a tile — lower values are blurrier and get dropped. |
Stain normalization
| Parameter | Type | Default | What it does |
|---|---|---|---|
normalize | boolean | true | Apply stain-color normalization. |
normalization_method | enum | reinhard | Method: reinhard, macenko, or vahadane. |
Output options
| Parameter | Type | Default | What it does |
|---|---|---|---|
output_format | enum | png | Tile image format: png, tiff, or jpg. |
save_masks | boolean | true | Upload the tissue mask (tissue-detection workflow). |
save_features | boolean | true | Upload the feature matrix (feature-extraction workflow). |
Accepted slide formats
The processor opens SVS, NDPI, and pyramidal TIFF via histolab/OpenSlide, and standard PNG/JPG images via PIL as a fallback. Very large slides are read region by region during tiling.
Execution profile
Pathology analysis runs on GPU (A10G), asynchronously, and can take 10–60 minutes depending on slide size and workflow. Input is a file path (the slide); outputs are JSON plus downloadable artifacts.
Read next
Was this page helpful?