CSCL

Cross-Stain Contrastive Learning for Paired Immunohistochemistry and Histopathology Slide Representation ( CSCL)

We provide bash script for training and inference on Our dataset Before running the code, ensure you have downloaded these dataset and preprocess following CONCH You can download our dataset from

HER2
PGR
ER
KI67

The list of useful parameters is as follows:

• local_dir: this is the path where you put your '.pt' files for patch features.
• model_dir: the location of CS or CL model weights.
• slide_embedding_pkl: the way to slide embedding.
• label_path: the way to test dataset label CSV file.

Train Adapter (CL)

cd CL
cd ./bin
bash ../scripts/train_adapter.sh

Using Adapter for Patch Embedding

bash extract_Adapter_embedding.sh

Train CS

cd CS
cd ./bin
bash ../scripts/launch_pretrain_withStainEncodings.sh

Using CS for Slide Embedding

cd ./bin
python extract_slide_embeddings.py

Inference

python run_linear_probing.py