rendeirolab/LazySlide

What it solves

LazySlide addresses the complexity and lack of interoperability in whole slide image (WSI) analysis. It provides a standardized framework that allows researchers to move from raw histological slides to deep learning-ready data without having to build custom, fragmented pipelines for preprocessing and feature extraction.

How it works

The framework integrates with the scverse ecosystem (using SpatialData) to provide a consistent API for digital pathology. It handles the heavy lifting of WSI processing through a pipeline that typically includes tissue segmentation (finding the tissue), tessellation (breaking the image into smaller tiles), and feature extraction using pre-trained models. It also provides PyTorch dataloaders to feed this processed data directly into machine learning models.

Who it’s for

It is designed for computational biologists, pathologists, and AI researchers who need to analyze large-scale histological images and integrate them with other modalities like transcriptomics or genomics.

Highlights

• Foundation Model Support: Native integration with state-of-the-art models such as UNI, CONCH, Gigapath, and Virchow for captioning and zero-shot classification.
• scverse Compatibility: Fully compatible with tools like scanpy, anndata, and squidpy.
• Multimodal Integration: Ability to combine histological data with textual annotations, genomics, and transcriptomics.
• Scalable Pipeline: Efficiently handles high-throughput analysis of large WSIs.