X-ray Chest

CheXpert

Name: CheXpert: CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison
Published: 2019

CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison

224,316 chest radiographs of 65,240 Stanford Hospital patients (2002-2017) with 14 report-derived labels that include an "uncertain" state, plus a radiologist-labelled validation set. Non-commercial research use only.

Go to dataset Edit Report

CheXpert is a chest radiograph collection from Stanford Hospital released by the Stanford Machine Learning Group and distributed by the Stanford AIMI Center. Its main contribution is a rule-based labeler that reads radiology reports and marks each of 14 observations as positive, negative or uncertain. It is one of the standard benchmarks for multi-label chest X-ray classification.

Composition

The dataset holds 224,316 radiographs of 65,240 patients, taken between October 2002 and July 2017 in inpatient and outpatient settings. Each patient appears in only one split:

Split	Patients	Studies	Images
Training	64,540	187,641	223,414
Validation	200	200	234
Test	500	500	668

Each image is a frontal (AP or PA) or lateral view. Age in years, biological sex and the view are given per image. Images are distributed as 8-bit JPEG files, in a full-resolution and a downsampled version. The datasheet reports that the X-ray device is not recorded.

Annotations

The 14 observations are No Finding, Enlarged Cardiomediastinum, Cardiomegaly, Lung Opacity, Lung Lesion, Edema, Consolidation, Pneumonia, Atelectasis, Pneumothorax, Pleural Effusion, Pleural Other, Fracture and Support Devices. Training labels come from the automatic labeler run on the reports. Validation labels are the majority vote of three board-certified radiologists looking at the images, and test labels the majority vote of five. This index maps the findings that have a vocabulary term; condition counts in the statistics include positive labels only.

Known limitations

Training labels are extracted from text and contain errors; uncertain labels are frequent for some findings (e.g. consolidation).
Single institution; about half of the images come from 15% of the patients.
Very few children are included.
The research use agreement forbids commercial use, redistribution and derivative works. A paid commercial license is offered separately.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 65,240 subjects.

Split

subjects

Training 64,540 99%
Test 500 <1%
Validation 200 <1%

License and access

Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer

Access

Free registration

Create an account and accept terms online

Access page

Stanford University School of Medicine Dataset Research Use Agreement (AIMI)

Free for personal, non-commercial research only. You may not share the data or the download link, may not create derivative works, and must not try to re-identify patients. Stanford sells a separate commercial license.

Original license text Version read: Text shown in the Stanford AIMI shared-datasets portal, read 2026-10-07 (no version number given) Checked 2026-10-07

What you can do

No
Not stated
No
Not stated

What you can share

No
No
Share trained models Not stated

What you must do

Yes
Share alike No
Yes
Ethics approval No
Manuscript review No
Release code No
Return results No
No

Limits

Yes
Location limits No

Commercial license: Yes Companies can apply for a one-year license per dataset, reviewed by a Stanford committee. The page lists an annual fee of USD 70,000 per dataset for agreements contracted in FY25. Details

Citation

Irvin J, Rajpurkar P, Ko M, et al. CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison. AAAI 2019. https://arxiv.org/abs/1901.07031. Dataset DOI: https://doi.org/10.71718/y7pj-4v93

All numbers

Every number on this page, as stored in stats.csv, with its source.

Measure	Breakdown	Value	Source
Subjects	total	65,240	irvin2019 Abstract
Images	total	224,316	irvin2019 Abstract
Subjects	split=train	64,540	garbin2021 Table I
Subjects	split=validation	200	garbin2021 Table I
Subjects	split=test The 2021 datasheet says the test set is not publicly available	500	garbin2021 p. 2 (Composition)
Studies	split=train	187,641	garbin2021 Table I
Studies	split=validation	200	garbin2021 Table I
Studies	split=test	500	irvin2019 p. 6 (Test Set)
Images	split=train	223,414	garbin2021 Table I
Images	split=validation	234	garbin2021 Table I
Images	split=test	668	garbin2021 p. 2 (Composition)
Subjects	split=train;sex=female	28,729	garbin2021 Table VII
Subjects	split=train;sex=male	35,811	garbin2021 Table VII
Subjects	split=validation;sex=female	94	garbin2021 Table VII
Subjects	split=validation;sex=male	106	garbin2021 Table VII
Images	split=train;sex=female	90,778	garbin2021 Table VII
Images	split=train;sex=male	132,636	garbin2021 Table VII
Images	split=validation;sex=female	106	garbin2021 Table VII
Images	split=validation;sex=male	128	garbin2021 Table VII
Studies	split=train;age=0-1 Age at the time of the study; MeSH age groups	3	garbin2021 Table VIII
Studies	split=train;age=13-18	653	garbin2021 Table VIII
Studies	split=train;age=19-44	34,450	garbin2021 Table VIII
Studies	split=train;age=45-64	69,059	garbin2021 Table VIII
Studies	split=train;age=65-79	52,934	garbin2021 Table VIII
Studies	split=train;age=80+	30,542	garbin2021 Table VIII
Studies	split=validation;age=13-18	1	garbin2021 Table VIII
Studies	split=validation;age=19-44	33	garbin2021 Table VIII
Studies	split=validation;age=45-64	75	garbin2021 Table VIII
Studies	split=validation;age=65-79	57	garbin2021 Table VIII
Studies	split=validation;age=80+	34	garbin2021 Table VIII
Studies	split=train;sex=female;age=0-1	3	garbin2021 Table VIII
Studies	split=train;sex=female;age=13-18	254	garbin2021 Table VIII
Studies	split=train;sex=male;age=13-18	399	garbin2021 Table VIII
Studies	split=train;sex=female;age=19-44	14,395	garbin2021 Table VIII
Studies	split=train;sex=male;age=19-44	20,055	garbin2021 Table VIII
Studies	split=train;sex=female;age=45-64	27,026	garbin2021 Table VIII
Studies	split=train;sex=male;age=45-64	42,033	garbin2021 Table VIII
Studies	split=train;sex=female;age=65-79	21,653	garbin2021 Table VIII
Studies	split=train;sex=male;age=65-79	31,281	garbin2021 Table VIII
Studies	split=train;sex=female;age=80+	14,846	garbin2021 Table VIII
Studies	split=train;sex=male;age=80+	15,696	garbin2021 Table VIII
Images	split=train;age=0-1 Age at the time of the study; MeSH age groups	3	garbin2021 Table VIII
Images	split=train;age=13-18	766	garbin2021 Table VIII
Images	split=train;age=19-44	41,696	garbin2021 Table VIII
Images	split=train;age=45-64	82,750	garbin2021 Table VIII
Images	split=train;age=65-79	62,433	garbin2021 Table VIII
Images	split=train;age=80+	35,766	garbin2021 Table VIII
Images	split=validation;age=13-18	2	garbin2021 Table VIII
Images	split=validation;age=19-44	38	garbin2021 Table VIII
Images	split=validation;age=45-64	88	garbin2021 Table VIII
Images	split=validation;age=65-79	64	garbin2021 Table VIII
Images	split=validation;age=80+	42	garbin2021 Table VIII
Images	split=train;sex=female;age=0-1	3	garbin2021 Table VIII
Images	split=train;sex=female;age=13-18	290	garbin2021 Table VIII
Images	split=train;sex=male;age=13-18	476	garbin2021 Table VIII
Images	split=train;sex=female;age=19-44	16,972	garbin2021 Table VIII
Images	split=train;sex=male;age=19-44	24,724	garbin2021 Table VIII
Images	split=train;sex=female;age=45-64	31,653	garbin2021 Table VIII
Images	split=train;sex=male;age=45-64	51,097	garbin2021 Table VIII
Images	split=train;sex=female;age=65-79	24,817	garbin2021 Table VIII
Images	split=train;sex=male;age=65-79	37,616	garbin2021 Table VIII
Images	split=train;sex=female;age=80+	17,043	garbin2021 Table VIII
Images	split=train;sex=male;age=80+	18,723	garbin2021 Table VIII
Images	split=validation;sex=female;age=13-18	2	garbin2021 Table VIII
Images	split=validation;sex=female;age=19-44	22	garbin2021 Table VIII
Images	split=validation;sex=male;age=19-44	16	garbin2021 Table VIII
Images	split=validation;sex=female;age=45-64	39	garbin2021 Table VIII
Images	split=validation;sex=male;age=45-64	49	garbin2021 Table VIII
Images	split=validation;sex=female;age=65-79	25	garbin2021 Table VIII
Images	split=validation;sex=male;age=65-79	39	garbin2021 Table VIII
Images	split=validation;sex=female;age=80+	18	garbin2021 Table VIII
Images	split=validation;sex=male;age=80+	24	garbin2021 Table VIII
Studies	split=train;condition=no_finding Positive label from the report labeler; uncertain labels not counted	16,627	irvin2019 Table 1
Studies	split=train;condition=cardiomegaly Positive only; 6,597 more uncertain	23,002	irvin2019 Table 1
Studies	split=train;condition=lung_opacity Positive only	92,669	irvin2019 Table 1
Studies	split=train;condition=pulmonary_edema Positive only	48,905	irvin2019 Table 1
Studies	split=train;condition=consolidation Positive only	12,730	irvin2019 Table 1
Studies	split=train;condition=pneumonia Positive only	4,576	irvin2019 Table 1
Studies	split=train;condition=atelectasis Positive only	29,333	irvin2019 Table 1
Studies	split=train;condition=pneumothorax Positive only	17,313	irvin2019 Table 1
Studies	split=train;condition=pleural_effusion Positive only	75,696	irvin2019 Table 1
Studies	split=train;condition=fracture Positive only	7,270	irvin2019 Table 1
Images	split=train;condition=no_finding Positive label from the report labeler; uncertain labels not counted	22,381	garbin2021 Table IV
Images	split=train;condition=cardiomegaly Positive only	27,000	garbin2021 Table IV
Images	split=train;condition=lung_opacity Positive only	105,581	garbin2021 Table IV
Images	split=train;condition=pulmonary_edema Positive only	52,246	garbin2021 Table IV
Images	split=train;condition=consolidation Positive only	14,783	garbin2021 Table IV
Images	split=train;condition=pneumonia Positive only	6,039	garbin2021 Table IV
Images	split=train;condition=atelectasis Positive only	33,376	garbin2021 Table IV
Images	split=train;condition=pneumothorax Positive only	19,448	garbin2021 Table IV
Images	split=train;condition=pleural_effusion Positive only	86,187	garbin2021 Table IV
Images	split=train;condition=fracture Positive only	9,040	garbin2021 Table IV
Images	split=validation;condition=no_finding Radiologist consensus labels	38	garbin2021 Table V
Images	split=validation;condition=cardiomegaly	68	garbin2021 Table V
Images	split=validation;condition=lung_opacity	126	garbin2021 Table V
Images	split=validation;condition=pulmonary_edema	45	garbin2021 Table V
Images	split=validation;condition=consolidation	33	garbin2021 Table V
Images	split=validation;condition=pneumonia	8	garbin2021 Table V
Images	split=validation;condition=atelectasis	80	garbin2021 Table V
Images	split=validation;condition=pneumothorax	8	garbin2021 Table V
Images	split=validation;condition=pleural_effusion	67	garbin2021 Table V

Sources

The keys used in the table above.

irvin2019 Irvin et al. 2019, CheXpert (AAAI), arXiv 1901.07031 paper
garbin2021 Garbin et al. 2021, Structured dataset documentation - a datasheet for CheXpert, arXiv 2105.03020 paper
aimi-chexpert Stanford AIMI dataset page website

At a glance

Subjects: 65,240
Images: 224,316

Conditions: Atelectasis, Cardiomegaly, Pulmonary consolidation, Bone fracture, Pulmonary opacity, No finding, Pleural effusion, Pneumonia, Pneumothorax, Pulmonary edema
Tasks: Classification
Released: 2019
Format: JPEG
Countries: United States
DOI: 10.71718/y7pj-4v93
Creators: Stanford Machine Learning Group; Stanford Center for Artificial Intelligence in Medicine and Imaging (AIMI)

License at a glance

Commercial use No
Train ML models Not stated
Share the data No
Share trained models Not stated
Sign an agreement Yes
Ethics approval No

All 17 rules with quotes

Our interpretation, not legal advice. Disclaimer

Checked against its sources on 2026-10-07 by claude-research.

MIMIC-CXR
377,110 chest radiographs in DICOM format from 227,835 studies of 65,379 patients seen at the Beth Israel Deaconess Medical Center emergency department (2011-2016), each study paired with its free-text radiology report.
NIH ChestX-ray14
112,120 frontal chest radiographs (PNG) of 30,805 patients from the NIH Clinical Center, with 14 thoracic disease labels text-mined from reports, age, sex, view, an official patient-level split and ~1,000 bounding boxes.

Overview