Detecting Shortcuts in Medical Images - A Case Study in Chest X-rays

by Amelia Jiménez-Sánchez, Dovile Juodelyte, Bethany Chamberlain, Veronika Cheplygina

This repository provides a PyTorch implementation of our work accepted at ISBI 2023 -> [PDF] [arXiv]

Overview

Data-centric approaches, bias assessment, and validation are increasingly important as datasets get larger, but are still understudied in medical imaging. We review the literature and present a validation study on detecting shortcuts in chest X-rays. Our systematic experiments on two large benchmarks generalize earlier findings which show overoptimistic and biased performance. We share our code and a set of non-expert drain labels for CheXpert dataset under the preprocess folder.

Usage

1. Cloning the repository

$ git clone https://github.com/ameliajimenez/shortcuts-chest-xray.git
$ cd shortcuts-chest-x-ray/

2. Preprocessing: create development subsets

Detailed steps under preprocess folder.

3. Training, testing & visualizations

Detailed steps under bin folder.

Citation

If this work is useful for your research, please cite our paper:

@INPROCEEDINGS{10230572,
  author={Jiménez-Sánchez, Amelia and Juodelyte, Dovile and Chamberlain, Bethany and Cheplygina, Veronika},
  booktitle={2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI)}, 
  title={Detecting Shortcuts in Medical Images - A Case Study in Chest X-Rays}, 
  year={2023},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/ISBI53787.2023.10230572}}

Acknowledgments

Our repository is based on jhealthcare/CheXpert and purrlab/hiddenfeatures-chestxray. We thank Kasper Thorhauge Grønbek and Andreas Skovdal for early discussions and providing the labels used in our experiments.