alterlab-histolab

Files (9)

• SKILL.md
• evals.json · evals/
• best_practices.md · references/
• filters_preprocessing.md · references/
• slide_management.md · references/
• tile_extraction.md · references/
• tissue_masks.md · references/
• visualization.md · references/
• workflows.md · references/

SKILL.md (first lines)

# Histolab

## Overview

Histolab is a Python library for processing whole slide images (WSI) in digital pathology. It automates tissue detection, extracts informative tiles from gigapixel images, and prepares datasets for deep learning pipelines. The library handles multiple WSI formats, implements sophisticated tissue segmentation, and provides flexible tile extraction strategies.

## When to Use This Skill

Use histolab for lightweight WSI tile pipelines: tissue detection, building tile datasets for ML training, H&E stain handling, and quick tile-based analysis of histopathology slides. For advanced spatial proteomics, multiplexed imaging, or full deep-learning pathology pipelines, use `alterlab-pathml` instead.

### Does NOT Trigger

| Scenario | Use Instead |
|----------|-------------|
| End-to-end computational pathology, nucleus segmentation, multiplexed imaging (CODEX/Vectra) | `alterlab-pathml` |
| Training the downstream classifier on the extracted tiles | `alterlab-pytorch-lightning` |
| DICOM whole-slide or radiology images rather than SVS/NDPI | `alterlab-pydicom`, `alterlab-imaging-data-commons` |
| Spatial transcriptomics on tissue sections | `alterlab-squidpy-spatial` |
| Managing/versioning the slide collection itself | `alterlab-omero`, `alterlab-lamindb` |
| Single-cell RNA-seq analysis of an `.h5ad` (QC, normalisation, Leiden clustering, marker genes) | `alterlab-scanpy` |

**Hand-off rule.** When the request matches a row above, open your answer by naming that skill by its exact id (for example `alterlab-pathml`) and saying in one line why it fits better, then stop applying this skill's procedure. Name it even when you also need to ask for missing input. If only part of the request is out of scope, name the skill for that part and do only the in-scope part. A brief general pointer is fine; the named skill does the work.

## Installation

```bash
# histolab 0.7.0 needs Python >= 3.8, < 3.12 and numpy 1.x — give it its own env
uv venv --python 3.11 .venv-histolab
uv pip install --python .venv-histolab "histolab==0.7.0" openslide-bin
```

Two constraints that bite in a 2026 toolchain:

1. **Pinned scientific stack.** 0.7.0 (Feb 2024, still the latest release) requires
   `numpy <= 1.24.4`, `scipy < 1.10.1`, `scikit-image < 0.19.4` and
   `openslide-python 1.3.1`. It cannot share an environment with a numpy 2.x / Python
   3.12+ project — resolve this with a separate venv rather than by loosening pins.
2. **OpenSlide is a C library**, not bundled with the wheel. The simplest fix is the
   `openslide-bin` wheel (prebuilt binaries); otherwise `brew install openslide` on macOS
   or `apt install libopenslide0` on Debian/Ubuntu. Without it, `import histolab.slide`
   fails with `Couldn't locate OpenSlide dylib`.

The examples below target histolab 0.7.0; the API differs in older releases.

## Core Workflow

1. **Load** the slide with `Slide(path, processed_path=...)` and inspect dimensions/levels.
2. **Detect tissue** with a mask (`TissueMask` or `BiggestTissueBoxMask`).
3. **Preview** tile locations with `tiler.locate_tiles(slide)` before committing.
4. **Extract** tiles with one of three tilers (Random/Grid/Score).

Minimal example:

```python
from histolab.slide import Slide
from histolab.tiler import RandomTiler

slide = Slide("slide.svs", processed_path="output/")
# n_tiles, level, seed are CONSTRUCTOR args — not args to locate_tiles/extract.
tiler = RandomTiler(tile_size=(512, 512), n_tiles=100, level=0, seed=42)
tiler.locate_tiles(slide)   # preview locations on the thumbnail first
tiler.extract(slide)        # writes PNGs into processed_path
```

**API gotcha (histolab 0.7.0):** `locate_tiles()` and `extract()` take only
`slide`, an optional `extraction_mask`, and logging/styling kwargs — they do
**not** accept `n_tiles`. Set `n_tiles` (and `seed`, `level`, `tile_size`,
`check_tissue`, `tissue_percent`) on the tiler constructor. The
`extraction_mask` is passed to `extract()`/`locate_tiles()`, never to the
constructor.

Full copy-pasteable pipelines (quick start, 5 end-to-end workflows, and per-capability examples) live in `references/workflows.md`.

## Core Capabilities

### 1. Slide Management

Load, inspect, and work with WSI files (SVS, TIFF, NDPI, etc.): access metadata
(dimensions, magnification, properties), generate thumbnails, and work with
pyramidal/multi-level structures. Key class: `Slide`.

See `references/slide_management.md` for slide initialization, built-in sample
datasets (`prostate_tissue`, `ovarian_tissue`, `breast_tissue`, `heart_tissue`,
`aorta_tissue`, plus pen-marked and IHC samples), pyramid levels, and
multi-slide processing.

### 2. Tissue Detection and Masks

Automatically identify tissue regions and filter background/artifacts. Key
classes: `TissueMask` (all tissue regions), `BiggestTissueBoxMask` (bounding box
of largest region — the default), and `BinaryMask` (base class for custom masks).

Choosing a mask:
- `TissueMask`: multiple tissue sections, comprehensive analysis
- `BiggestTissueBoxMask`: single main section, exclude artifacts (default)
- Custom `BinaryMask`: specific ROI, exclude annotations, custom segmentation

See `references/tissue_masks.md` for how detection filters work, visualizing
masks with `locate_mask()`, and custom rectangular / annotation-exclusion masks.

### 3. Tile Extraction

Extract smaller regions from large WSI using one of three strategies:

- **RandomTiler** — fixed number of randomly positioned tiles. Best for sampling
  diverse regions, exploration, training data. Key params: `n_tiles`, `seed`.
- **GridTiler** — systematic grid across tissue. Best for complete coverage,
  spatial analysis, reconstruction. Key param: `pixel_overlap`.
- **ScoreTiler** — top-ranked tiles by scoring function. Best for informative
  regions, quality-driven selection. Key param: `scorer` (NucleiScorer,
  CellularityScorer, custom).

Common parameters: `tile_size`, `level` (0 = highest res), `check_tissue`,
`tissue_percent` (default 80%), `extraction_mask`. **Always preview with
`locate_tiles()` before extracting.**

See `references/tile_extraction.md` for scorers, reporting, and advanced
(multi-level, hierarchical) extraction patterns.

From AlterLab-IEU/AlterLab-Academic-Skills, Apache-2.0.

More from AlterLab-IEU/AlterLab-Academic-Skills