RPMIL

This repository provides the code for the PyTorch implementation of RPMIL: Rethinking Uncertainty-Aware Probabilistic Multiple Instance Learning for Whole Slide Pathology Diagnosis, IJCAI 2025.

For running on a local machine

• Linux
• Python 3.10.0, Torch 2.5.1
• Nvidia GeForce RTX 3090 GPU

Whole Slide Image Data

Clink Camelyon16 and TCGA-NSCLC to download original WSIs.

Data Preprogressing

We use CLAM to preprocess WSIs.

git clone https://github.com/mahmoodlab/CLAM.git
cd CLAM
conda env create -f env.yml
conda activate clam_latest

Patching

--preset bwh_biopsy.csv for C16 (It's the preset parameters officially provided by CLAM),
--preset preprocess_tcga_nsclc.csv for TCGA-NSCLS (It's the customized parameters).

# for C16
python create_patches_fp.py --source DATA_DIRECTORY --save_dir RESULTS_DIRECTORY --patch_size 512 \
--step_size 512 --preset bwh_biopsy.csv --seg --patch
# for TCGA-NSCLC
python create_patches_fp.py --source DATA_DIRECTORY --save_dir RESULTS_DIRECTORY --patch_size 512 \
--step_size 512 --preset preprocess_tcga_nsclc.csv --seg --patch

Feature Extraction

# for C16
CUDA_VISIBLE_DEVICES=0 python extract_features_fp.py --data_h5_dir DIR_TO_COORDS --data_slide_dir \
DATA_DIRECTORY --csv_path CSV_FILE_NAME --feat_dir FEATURES_DIRECTORY --batch_size 512 --target_patch_size=224 --slide_ext .tif
# for TCGA-NSCLC
CUDA_VISIBLE_DEVICES=0 python extract_features_fp.py --data_h5_dir DIR_TO_COORDS --data_slide_dir \
DATA_DIRECTORY --csv_path CSV_FILE_NAME --feat_dir FEATURES_DIRECTORY --batch_size 512 --target_patch_size=224 --slide_ext .svs

Creating the virtual environment

conda create -n rpmil python=3.10.0
conda activate rpmil

Getting the code

git clone https://github.com/bravePinocchio/rpmil
cd rpmil
pip install -r requirements.txt

Running the code

CUDA_VISIBLE_DEVICES=0 python main.py