slideflow
Deep learning library for digital pathology and whole-slide image analysis.
Open-source Python library for digital pathology, model development, and whole-slide image analysis.
Slideflow is a deep learning library tailored for digital pathology, bridging advanced machine learning with clinical research needs. Medical researchers and data scientists use it to process whole-slide images and train predictive models.
What it is
It provides customizable training pipelines, slide processing, and stain normalization tools that handle weakly and strongly-supervised labels. You can build models using multiple-instance learning, self-supervised learning, generative adversarial networks, and built-in foundation models.
Who it's for
It suits computational pathologists and medical AI researchers with Python experience who need end-to-end slide analysis workflows. It does not suit clinicians without programming skills or those looking for a turnkey medical viewing application without coding.
Key features
Slide Processing Toolkit
Extracts tiles from whole-slide images using cuCIM or Libvips with adjustable resolution and magnification.
Stain Normalization
Applies built-in stain normalization routines to standardize histology appearance across different scanners.
Multiple-Instance Learning
Supports MIL architectures like CLAM for slide-level prediction using patient annotations.
Explainability Maps
Generates heatmaps, mosaic maps, saliency maps, and synthetic histology via StyleGAN.
Up and running in 3 steps
Our honest verdict
What's great
- Supports both PyTorch and Tensorflow backends depending on your environment preference.
- Includes extensive slide processing and tiling utilities with RAM disk buffering options.
- Offers advanced explainability tools including heatmaps and layer activation analysis.
- Integrates with popular pre-trained foundation models like UNI, HistoSSL, and GigaPath.
Where it falls short
- Requires significant Python and machine learning expertise to set up and configure properly.
- Demands high-performance hardware, specifically GPUs with CUDA support, to train models efficiently.
- Some advanced add-on models like CLAM and StyleGAN require installing non-commercial packages.
- Slide reading performance depends on properly configured external dependencies like cuCIM or Libvips.
Pricing & license
Free under the Apache-2.0 license, with optional non-commercial add-ons available. The only cost is running it yourself on your own GPU hardware.
Open remote AI-training roles
874 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Which deep learning frameworks does Slideflow support?
Slideflow supports both PyTorch and Tensorflow, defaulting to PyTorch if both are available and configurable via the SF_BACKEND environment variable.
How are whole-slide images read?
It uses cuCIM by default for fast reading, with an optional Libvips backend to support formats like .scn, .mrxs, .ndpi, .vms, and .vmu.
