HistoStainAlign is the official repository for the paper:
Cross-Modality Learning for Predicting IHC Biomarkers from H&E-Stained Whole-Slide Images.

HistoStainAlign enables cross-modality prediction of immunohistochemistry (IHC) biomarkers from hematoxylin and eosin (H&E) stained whole-slide images (WSIs), leveraging deep learning and domain adaptation.

• Tile Embedding Extraction: Process WSIs to extract tile-level embeddings using Gigapath.
• Flexible Training: Train models with customizable classification heads.
• Slide-Level Embedding Generation: Aggregate tile embeddings for slide-level analysis.
• Evaluation Tools: Includes linear probe scripts for benchmarking embeddings.

1. Clone the repository

   git clone https://github.com/BMIRDS/HistoStainAlign.git
   cd HistoStainAlign

2. Install dependencies

   • All Python dependencies are listed in requirements.txt:

     pip install -r requirements.txt

   • Additionally, install Gigapath and its dependencies:

     git clone https://github.com/prov-gigapath/prov-gigapath.git
     cd prov-gigapath
     pip install -r requirements.txt

The workflow consists of four main steps:

1. Extract tile embeddings
   • 00_extract_tile_embeds.py: Use Gigapath’s tile encoder to generate embeddings.
2. Train the model
   • 01_train_model_with_classification_head.py: Train using the HistoStainAlign framework.
3. Generate slide-level embeddings
   • 02_generate_slide_embeddings.py: Aggregate tile embeddings for each WSI.
4. Evaluate with linear probe
   • 03_run_linear_probe.py: Assess slide embeddings using linear probing.

Refer to comments within each script for detailed usage and parameter options.

To use this repository, you will need access to appropriately formatted WSIs and IHC biomarker labels. Data preparation steps can be adapted from the scripts provided.

Contributions, issues, and feature requests are welcome!
Feel free to open an issue or submit a pull request.

Distributed under the GPL-3.0 License. See LICENSE for more information.

If you use this code or its ideas in your research, please cite:

@article{HistoStainAlign2024,
  title={Cross-Modality Learning for Predicting IHC Biomarkers from H&E-Stained Whole-Slide Images},
  author={Your Authors},
  journal={arXiv preprint arXiv:2506.15853},
  year={2024}
}