ISPY2 Dataset Curation for Breast Cancer pCR Prediction

Implementation of the ISPY2 dataset curation pipeline for breast cancer pathologic complete response (pCR) prediction from our publication "Graph Representation Learning of Longitudinal Medical Imaging Trajectories for Treatment Response Prediction".

Example Usage

1. Create a Python file with the following content (e.g. ispy2_pcr_dataset.py)

from ISPY2_pCR_dataset import download_and_preprocess

if __name__ == "__main__":

    download_dir = "/path/to/output/directory"
    manifest_file_path = "/path/to/manifest_xlsx/file" # can be found at https://github.com/hannesk95/ISPY2-pCR-dataset/tree/main/data
    clinical_file_path = "/path/to/clinical_xlsx/file" # can be found at https://github.com/hannesk95/ISPY2-pCR-dataset/tree/main/data

    download_and_preprocess(
        download_dir=download_dir,
        manifest_file_path=manifest_file_path,
        clinical_file_path=clinical_file_path
        )

2. Start a tmux session, run the Python script, and enjoy a coffee (or several) this will take a while!

$ tmux
$ python ispy2_pcr_dataset.py

3. Once the script has finished, you'll find the final dataset at

$ cd download_dir/processed_data

Citation

@article{kiechle2026graph,
title={Graph Representation Learning of Longitudinal Medical Imaging Trajectories for Treatment Response Prediction},
author={Kiechle, Johannes and Osuala, Richard and Lang, Daniel M and Fischer, Stefan M and Jan{\'\i}{\v{c}}kov{\'a}, Ivana and Lekadir, Karim and Schnabel, Julia A and Peeken, Jan C},
journal={arXiv preprint arXiv:2607.04912},
year={2026}
}