Skip to content
Open Imaging Index
X-ray Chest

NIH ChestX-ray14

NIH Clinical Center Chest X-ray Dataset (ChestX-ray8 / ChestX-ray14)

Frontal chest radiographs with disease labels mined from reports, one of the first chest X-ray sets large enough for deep learning, used for multi-label classification and weakly supervised localization. 112,120 images of 30,805 patients with 14 labels, age, sex, view and about 1,000 boxes.

Go to dataset Edit Report

Overview

NIH ChestX-ray14 is a set of frontal chest radiographs released by the NIH Clinical Center in 2017. It grew out of the ChestX-ray8 work presented at CVPR 2017 and became one of the first chest X-ray collections large enough for deep learning. It is mostly used for multi-label classification and weakly supervised localization of thoracic findings.

Composition

The release contains 112,120 frontal images (PA or AP) of 30,805 patients, stored as 1024 x 1024 PNG files. The README states that the images are taken from the Clinical Center PACS and make up about 60% of all frontal chest radiographs there. Many patients have follow-up images. A metadata table gives, per image, the labels, follow-up number, patient id, age, sex, view position and the original size and pixel spacing. The ages and follow-up numbers were corrected in an April 2020 update of this table.

The creators provide a split at patient level: a training/validation list and a test list. All images of one patient are in the same list. The numbers on this page count the official training/validation list as train.

Annotations

Fourteen findings were text-mined from the radiology reports with natural language processing: atelectasis, cardiomegaly, effusion, infiltration, mass, nodule, pneumonia, pneumothorax, consolidation, edema, emphysema, fibrosis, pleural thickening and hernia. Images without any of them are labelled "No Finding". An image can have several labels. About 1,000 images also have bounding boxes for one finding each. The reports themselves are not released.

Known limitations

  • Labels come from reports, not from reading the images; the creators estimate their accuracy at over 90%.
  • "No Finding" is not the same as normal: such images can show other diseases or uncertain findings.
  • Only PNG images are available, without the original DICOM headers.
  • Single US hospital; scanner and acquisition details are not documented.
  • Counts on this page were computed from the public metadata file.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 30,805 subjects.

Sex

  • Female 14,175 46%
  • Male 16,630 54%

Patient Sex column, Patient ID column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Condition

Groups can overlap

  • No finding 24,907 81%
  • Pulmonary infiltrates 8,035 26%
  • Atelectasis 4,981 16%
  • Pleural effusion 4,275 14%
  • Pulmonary nodule 3,394 11%
  • Pulmonary mass 2,568 8%
  • Pulmonary consolidation 2,150 7%
  • Pleural thickening 2,006 7%

Finding Labels column, Patient ID column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Condition by view

Images

Reported cross table. A dot marks a cell the source does not give.

PosteroanteriorAnteroposterior
No finding39,30221,059
Pulmonary infiltrates9,35310,541
Pleural effusion6,5896,728
Atelectasis5,7285,831
Pulmonary nodule4,1772,154
Pulmonary mass3,5672,215
Pneumothorax3,4071,895
Pulmonary consolidation1,5213,146
Pleural thickening2,418967
Cardiomegaly1,5631,213
Pulmonary emphysema1,4991,017
Pulmonary edema2762,027
Pulmonary fibrosis1,408278
Pneumonia630801
Hernia19235

View Position column, Finding Labels column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Sex by view

Images

Reported cross table. A dot marks a cell the source does not give.

PosteroanteriorAnteroposterior
Female30,16718,613
Male37,14326,197

Patient Sex column, View Position column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Age

mean 45.5 ± 16.7, median 47, range 0 to 95

0
2,000
4,000
6,000
7,117
0-910-1920-2930-3940-4950-5960-6970-7980-8990+

Patient Age column, Patient ID column, Follow-up # column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Split

  • Training 28,008 91%
  • Test 2,797 9%

test_list.txt / train_val_list.txt, Patient ID column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

View

  • Posteroanterior 28,868 94%
  • Anteroposterior 9,060 29%

View Position column, Patient ID column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Condition by split

Images

Reported cross table. A dot marks a cell the source does not give.

TrainingTest
No finding50,5009,861
Pulmonary infiltrates13,7826,112
Pleural effusion8,6594,658
Atelectasis8,2803,279
Pulmonary nodule4,7081,623
Pulmonary mass4,0341,748
Pneumothorax2,6372,665
Pulmonary consolidation2,8521,815
Pleural thickening2,2421,143
Cardiomegaly1,7071,069
Pulmonary emphysema1,4231,093
Pulmonary edema1,378925
Pulmonary fibrosis1,251435
Pneumonia876555
Hernia14186

Finding Labels column, test_list.txt / train_val_list.txt in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

Age by sex

Reported cross table. Missing cells were not published (fewer than 10 or not reported).

Female Male
  • 109
    80-89
    171
  • 582
    70-79
    1,062
  • 1,888
    60-69
    2,759
  • 3,312
    50-59
    3,805
  • 3,173
    40-49
    3,172
  • 2,417
    30-39
    2,523
  • 1,765
    20-29
    1,963
  • 651
    10-19
    815
  • 272
    0-9
    354

Patient Age column, Patient Sex column, Patient ID column, Follow-up # column in Data_Entry_2017_v2020.csv with test_list.txt and train_val_list.txt (public Box folder, use unrestricted per FAQ), counted per image and per unique Patient ID; joint cells below 10 dropped

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
Open download

Download without an account

Access page

NIH Clinical Center ChestX-ray dataset terms (FAQ)

Not a formal license. The NIH Clinical Center's FAQ states that use is unrestricted, but asks users to link to the original download site, acknowledge the NIH Clinical Center and cite the CVPR 2017 paper.

Original license text Version read: FAQ_CHESTXRAY.pdf as updated 2017-12-14 (per LOG_CHESTXRAY.pdf), read 2026-10-07 Checked 2026-10-07

What you can do

  • Yes
  • Yes
  • Yes
  • Yes
  • Yes

What you can share

  • Not stated
  • Yes
  • Share trained models Not stated

What you must do

  • Yes
  • Share alike No
  • No
  • No
  • Manuscript review No
  • Release code No
  • No
  • Delete after use No

Limits

  • No re-identification Not stated
  • Location limits No

Citation

Wang X, Peng Y, Lu L, Lu Z, Bagheri M, Summers RM. ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases. IEEE CVPR 2017, pp. 3462-3471.

@InProceedings{wang2017chestxray,
  author = {Wang, Xiaosong and Peng, Yifan and Lu, Le and Lu, Zhiyong and Bagheri, Mohammadhadi and Summers, Ronald},
  title = {ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases},
  booktitle = {2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages = {3462--3471},
  year = {2017}
}

Sources

Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.