pathml

Preprocess whole-slide images by loading 160+ formats, generating standardized tiles, and applying modular pipeline transforms for tissue detection, stain normalization, and nucleus detection. Store the processed tiles, masks, and metadata in HDF5 format for ML-ready batch processing.

Process multiparametric CODEX or MERFISH imaging data by collapsing multi-cycle runs, segmenting cells, and quantifying marker expression. The pipeline outputs the quantified spatial data directly into AnnData format for downstream analysis.

Normalize H&E stains by composing a reproducible preprocessing pipeline with stain normalization and denoising transforms. This pipeline handles unified pyramid and ROI access across 160+ slide formats to ensure consistent H&E inputs.

Spatial graph construction is supported by building cell, tissue, and spatial graphs from segmented nuclei. These graphs enable spatial analysis and can be used directly for ML training and inference workflows like HACTNet.

Manage large tiles and masks for HoVer-Net training by storing all features, masks, and metadata in HDF5. This format enables fast tile-based iteration and scalable batch processing for computational pathology modeling.

Load SVS and NDPI files by accessing image pyramids, metadata, and regions of interest through a unified whole-slide image loader. The tool then writes the standardized tiles and masks into an HDF5 dataset for ML pipelines.