Gradient Biotech

Slide quantification

Run the full computational pathology pipeline on a tissue slide — tile generation, tissue detection, cell segmentation, and spatial quantification with interactive viewer review.

Research question

What is the tissue composition of this slide, how many cells are detected, and what spatial metrics characterize the tissue architecture?

Who this is for

  • Researchers new to computational pathology needing a reproducible end-to-end workflow
  • Labs migrating from QuPath or Fiji scripts to a web-based pipeline with run provenance
  • Core facilities processing region images or whole-slide scans at scale

Data requirements

DataRequiredPurpose
Image fileYesPNG, JPEG, TIFF, OME-TIFF, or WSI formats (SVS, NDPI, MRXS)
Modality and stain typeRecommendedMetadata for study organization and QC context
OpenSlide (for WSI)Local envRequired for SVS/NDPI/MRXS decoding

For local smoke tests, use a PNG region image or the synthetic fixture from Sample data.

Workflow

Create study → Upload slide → Generate tiles → Tissue detect → Segment cells → Spatial quantification

Step 1 — Create study and upload

Create a pathology study and upload a slide from the study home page. Select modality (whole-slide image or region image) and stain type (H&E, IHC, IF).

Step 2 — Generate tiles

On the slide detail page, click Generate tiles. Wait for the job to complete — the slide viewer loads with pan/zoom tile navigation. Tiles are required before segmentation on large slides.

Step 3 — Tissue detection

Click Detect tissue. Review tissue fraction, focus score, and detected regions in the Tissue QC panel. A tissue mask overlay appears in the slide viewer.

Step 4 — Segment cells

Click Segment cells (requires tiles). Review cell count, model name, and mean area in the Segmentation Summary panel. Cell overlay and the detected-cell table populate in the viewer.

Available models are threshold_baseline (connected components) and watershed (splits touching nuclei via distance-transform watershed); Cellpose and StarDist deep-learning integration is planned but not yet available.

For IHC slides, optionally run Color deconvolution first to separate the Hematoxylin/DAB stain channels and get a DAB-positive area fraction. After segmentation, you can also train an Object classifier from annotation-labeled regions to assign a class to every cell beyond the default morphology-based label.

Step 5 — Spatial quantification

Click Spatial quantification (requires segmentation). Review:

  • Overall cell density
  • Infiltration phenotype summary
  • Nearest-neighbor distance statistics
  • Region-level metrics

Export CSV and JSON artifacts from the metrics panel. To hand the processed slide off to QuPath, Fiji, or OMERO, run Export OME-TIFF for a pyramidal, self-describing file with pixel-size metadata.

Expected outputs

ArtifactContents
Tile manifestMulti-level pyramid for web viewing
tissue_mask.pngBinary tissue foreground
cell_overlay.pngSegmentation visualization
cells.jsonCell centroids, areas, confidence scores
density_heatmap.pngSpatial density map
cell_spatial_metrics.csvPer-cell spatial measurements
region_metrics.jsonCompartment-level statistics

Typical paths

Slide typeNotes
Region PNG/JPEGFastest path for demos and algorithm validation
OME-TIFF region scanStandard pathology format with embedded metadata
Full WSI (40× H&E)Tile-first architecture; requires OpenSlide locally

Related guides