Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

CheXpert

Link

https://stanfordmlgroup.github.io/competitions/chexpert/

Indexing

Keywords: CheXpert, chest radiograph, competition, AUC, bias, sex, race, pleural effusion, cardiomegaly, pneumothorax

Content: CH

RadLex: RID5352, RID34539

Organization(s)

Stanford ML Group

Comments

CheXpert is a large, publicly available chest radiograph dataset used within a competition for automated chest radiograph interpretation with leaderboard rankings based on AUC. In the referenced study context, it was used to probe potential demographic bias (sex and race) in a chest radiography foundation model and to assess performance on pleural effusion, cardiomegaly, and pneumothorax.

References

[1] . "CheXpert. Stanford ML Group". . . Available from: https://stanfordmlgroup.github.io/competitions/chexpert/

[2] Glocker B, Jones C, Roschewitz M, Winczek S. "Risk of bias in chest radiography deep learning foundation models". Radiol Artif Intell. .

Dataset

Motivation

Automated chest radiograph interpretation at the level of practicing radiologists to support workflow prioritization, clinical decision support, large-scale screening, and global population health initiatives (as stated on the competition page quoted in the article).

Sensitive data

Demographic attributes (sex and race) were analyzed for bias in related work using this dataset.