pathml

What problem does it solve?

PathML provides a comprehensive toolkit to streamline computational pathology workflows: loading a wide range of whole-slide image formats, performing preprocessing (e.g., stain normalization, tissue detection), constructing spatial graphs, training machine learning models, and handling multiparametric imaging data. This reduces manual, repetitive analysis and enables reproducible, scalable pathology research and workflows.

Core Features & Use Cases

• Load WSIs from 160+ formats with unified access to pyramids and metadata
• Build modular preprocessing pipelines (stain normalization, tissue detection, nucleus detection)
• Construct cellular and tissue graphs for spatial analysis
• Train and deploy ML models (e.g., HoVer-Net, HACTNet) on pathology data
• Analyze multiparametric imaging (CODEX/Vectra) and export results to AnnData
• Manage large datasets with HDF5 storage and tile-based workflows

Use Case: Load a repository of H&E slides, apply tissue detection and stain normalization, segment nuclei, build tissue graphs, and train a nucleus classifier at scale, all in a reproducible workflow.

Quick Start

• Install PathML: pip install pathml
• Load a slide and run a basic pipeline (tissue detection + stain normalization): from pathml.core import SlideData wsi = SlideData.from_slide("path/to/slide.svs") pipeline = Pipeline([TissueDetectionHE(), StainNormalizationHE(target='normalize', stain_estimation_method='macenko')]) pipeline.run(wsi)