Extract slide-level embeddings from whole-slide images using PRISM.

PRISM Embedder is a containerized algorithm for computing slide-level representations from whole-slide images using PRISM.

⚠️ Access Required
This algorithm depends on the gated models PRISM and Virchow.
You need to have requested and been granted access to both models on Hugging Face.

Requirements:

• Linux-based OS (e.g., Ubuntu 22.04)
• Python 3.10+
• Docker installed

git clone https://github.com/clemsgrs/prism-embedder.git
cd prism-embedder

Download the required model weights:

./download_weights.sh

Replace </path/to/your/slide.tif> and </path/to/your/mask.tif> with your actual file paths.
Replace </path/to/your/output/folder> with the path to a folder where you have write access.

docker pull waticlems/prism_embedder:latest
docker run --rm -it \
    -v </path/to/your/slide.tif>:/input/images/whole-slide-image/slide.tif:ro \
    -v </path/to/your/mask.tif>:/input/images/tissue-mask/mask.tif:ro \
    -v model:/opt/ml/model
    -v </path/to/your/output/folder>:/output \
    waticlems/prism_embedder:latest

Upon execution, the following files will be saved in /output:

• image-neural-representation.json: 1280-dimensional slide-level embedding (JSON)
• whole-slide-tiling-thumbnail.png: PNG image showing the tissue tiling layout

• add tSNE plot
• improve CLI customization (e.g., for spacing, resolution)