What problem does it solve?
PathML streamlines computational pathology by turning whole-slide histology and multiparametric imaging (e.g., CODEX, Vectra) into usable tiles, masks, cell/graph representations, and quantification outputs for downstream machine learning and spatial analysis.
Core Features & Use Cases
- Whole-slide loading across 160+ formats: Read WSI pyramids and extract regions efficiently across vendor/proprietary formats (e.g., SVS/NDPI/DICOM/OME-TIFF).
- Reproducible preprocessing pipelines: Compose transforms for tissue detection, stain normalization, artifact/white-space labeling, and nucleus/cell segmentation.
- Cell segmentation and multiparametric quantification: Segment nuclei/cells (e.g., DeepCell Mesmer) and quantify marker expression into AnnData for single-cell-style workflows.
- Spatial graphs for analysis and GNNs: Build cell/tissue graphs from instance masks with configurable connectivity (knn/radius/delaunay/contact) and extract morphological/intensity/edge features.
- Dataset organization and storage: Persist processed tiles and features using HDF5 with efficient hierarchical structures and tile caching patterns.
- Use case: Process a CODEX cohort end-to-end—collapse runs, segment cells, quantify markers, cluster/visualize in spatial context, then export features for graph-based modeling.
Quick Start
Use the attached WSI to generate tiles, run a preprocessing pipeline for tissue detection and stain normalization, and then load resulting masks and tiles for subsequent segmentation and analysis.
