Datasets-Issues

Supporting code for our paper Limitations of Public Chest Radiography Datasets for Artificial Intelligence: Label Quality, Domain Shift, Bias and Evaluation Challenges, by Amy Rafferty, Rishi Ramaesh and Ajitha Rajan.

Included scripts:

• Data pre-processing and undersampling
• Training scripts for each of the seven model architectures
• Evaluation scripts for both threshold-dependant and threshold-independant evaluations

NOTE: Scripts are currently set up for use with a GPU.

Datasets all publicly available:

• MIMIC-CXR: https://physionet.org/content/mimic-cxr-jpg/2.1.0/
• CheXpert: https://stanfordmlgroup.github.io/competitions/chexpert/
• Chest X-ray14: https://www.kaggle.com/datasets/nih-chest-xrays/data?select=README_CHESTXRAY.pdf
• PadChest: https://bimcv.cipf.es/bimcv-projects/padchest/