computational-pathology

Computational Pathology

Whole-slide image processing and slide-level modelling for cancer histopathology. Covers reading vendor formats with OpenSlide, the coordinate and resolution semantics that cause most WSI bugs, tissue detection, tile extraction, stain normalization, H&E colour deconvolution, pathology foundation models as tile encoders, multiple instance learning for slide-level prediction, cell segmentation and classification, spatial statistics on cell positions, and integration with molecular data.

When to Use This Skill

Activate when the user requests:

• Reading .svs, .ndpi, .mrxs, .scn, or other whole-slide formats
• OpenSlide, TiaToolbox, histolab, or slideio pipelines
• Tissue detection or background removal on a slide
• Tile or patch extraction for downstream modelling
• Stain normalization across slides or scanners
• H&E colour deconvolution, or separating hematoxylin from eosin
• Working at a target magnification or microns-per-pixel
• Tile embeddings from UNI, CONCH, Virchow, Prov-GigaPath, H-optimus or Phikon
• Multiple instance learning with CLAM, DSMIL, TransMIL or attention pooling
• Slide-level prediction from tile features
• Splitting a pathology cohort without leakage
• Cell or nucleus segmentation (StarDist, HoVer-Net, Cellpose, InstanSeg)
• Tumour region or tertiary lymphoid structure detection
• Spatial statistics on segmented cell positions
• Correlating morphology with matched expression or mutation data

Inputs

Data Type	Format	Source
Whole-slide image	.svs, .ndpi, .mrxs, .tiff, .scn	GDC (TCGA), CAMELYON, PANDA
Slide metadata	OpenSlide properties	embedded in the file
Annotations	GeoJSON, XML, .qpdata	QuPath, ASAP, pathologist review
Reference tile	RGB image	a slide chosen as the stain target

---

Environment

Versions verified 2026-08.

pip install openslide-python openslide-bin   # 1.4.6 and 4.0.1.2

openslide-python is a BINDING, not the library. Its only dependency is
Pillow, so `pip install openslide-python` alone gives you an import that
fails at load time with a missing-library error.

openslide-bin ships the compiled OpenSlide C library (4.0.1) as a wheel,
which is why the upstream README tells you to install both. Before it
existed you needed a system package (apt/brew/conda). If you inherit a
pipeline that documents `brew install openslide`, that still works; do not
mix the two in one environment.