MIMIC-CXR
MIMIC Chest X-ray Database
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.
Overview
MIMIC-CXR is a large collection of chest radiographs with the radiology reports written for them during routine care. It was built by the MIT Laboratory for Computational Physiology together with Beth Israel Deaconess Medical Center (BIDMC) in Boston and is widely used for chest X-ray classification, report generation and vision-language research.
Composition
The cohort consists of patients who had a chest radiograph in the BIDMC emergency department between 2011 and 2016. For these patients, all chest radiograph studies from the same period were included. Version 2 contains 65,379 patients, 227,835 studies and 377,110 DICOM images. A study usually holds a frontal and a lateral image and always one free-text report. Patients often have several studies, and dates are shifted per patient into the years 2100 to 2200 so that the order of studies is kept.
The data descriptor lists the examination types: about two thirds of the images come from "CHEST (PA AND LAT)" exams and about one third from "CHEST (PORTABLE AP)" exams. View counts per image are not published. Version 2.1.0 (2024) added de-identified provider identifiers for each study.
Acquisition
Images were exported from the hospital PACS in their original DICOM format. Burned-in text that could identify a patient was blacked out by an OCR-based algorithm, and DICOM headers were cleaned following the DICOM de-identification profiles. Reports were de-identified with rule-based and neural methods, with removed text replaced by three underscores.
Annotations
MIMIC-CXR itself ships images and reports only. Structured labels (e.g. CheXpert-style findings) and JPEG versions are distributed in the separate MIMIC-CXR-JPG project.
Known limitations
- Single hospital in the US, with an emergency-department-derived cohort.
- Image quality and positioning vary as in clinical practice; black boxes cover some image regions, and some orientation metadata is wrong.
- The de-identification code is not public.
- Access requires credentialing, training and a data use agreement that forbids sharing the data.
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
Identity check and ethics training (e.g. CITI) plus an agreement
PhysioNet credentialed account, CITI "Data or Specimens Only Research" training and a signed data use agreement.
PhysioNet Credentialed Health Data License 1.5.0
Research-only license for credentialed PhysioNet users. You need human-subjects and HIPAA training and must sign a data use agreement. You must not share the data, try to identify people, or skip releasing the code behind your publications.
What you can do
- Not stated
- Not stated
- Create derived data Not stated
- Conditional
What you can share
- No
- Not stated
- Share trained models Not stated
What you must do
- Not stated
- Share alike No
- Yes
- No
- Manuscript review No
- Yes
- No
- No
Limits
- Yes
- Location limits No
Citation
Johnson AEW, Pollard TJ, Berkowitz SJ, et al. MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports. Sci Data 6, 317 (2019). https://doi.org/10.1038/s41597-019-0322-0. Johnson A, Pollard T, Mark R, Berkowitz S, Horng S. MIMIC-CXR Database (version 2.1.0). PhysioNet (2024). https://doi.org/10.13026/4jqj-jw95
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total | 65,379 | johnson2019 Abstract |
| Studies | total One study = one radiology report with one or more images | 227,835 | johnson2019 Abstract |
| Images | total DICOM images | 377,110 | johnson2019 Abstract |
Sources
The keys used in the table above.
- johnson2019 Johnson et al. 2019, Scientific Data (MIMIC-CXR data descriptor) paper
- physionet-mimic-cxr MIMIC-CXR Database v2.1.0 on PhysioNet website