MIMIC-CXR Dataset (224px resolution) / Dataset MIMIC-CXR (resolución 224px)
English
Dataset Description
This folder contains a clean MIMIC-CXR subset prepared for the Medical AI Datathon. Images
are preprocessed chest X-rays stored as JPG files. The CSV files include study
metadata, demographic variables, CheXpert-style labels, and the full radiology
report in the report column.
Original dataset: https://physionet.org/content/mimic-cxr/
Structure
MIMIC-CXR/
├── images/
├── train.csv
├── valid.csv
├── test.csv
└── README.md
The image column contains only the image filename. To load an image, combine
the dataset path, images/, and the value in image.
Files
images/: chest X-ray JPG files.train.csv,valid.csv,test.csv: split files with image names, metadata, labels, and reports.README.md: this file.
Main Variables
image: image filename insideimages/.dicom_id,subject_id,study_id: identifiers for the image, patient, and study.split: split name.ViewPosition: radiographic view, for example AP or PA.Rows,Columns: image dimensions.StudyDate,StudyTime: study timing metadata.race,sex,age: demographic variables.race_label,sex_label: encoded demographic variables.report: full radiology report text.
Label columns:
AtelectasisCardiomegalyConsolidationEdemaEnlarged CardiomediastinumFractureLung LesionLung OpacityNo FindingPleural EffusionPleural OtherPneumoniaPneumothoraxSupport Devices
For these labels, 1 means present, 0 means absent, -1 means uncertain,
and an empty value means no label is available.
Possible Tasks
- Multi-label chest X-ray classification.
- Report-aware or multimodal modeling using the
reportcolumn. - Subgroup or fairness analysis using metadata such as age, sex, and race.
Loading Example
from pathlib import Path
import pandas as pd
from PIL import Image
root = Path("PATH-TO-DATASET/MIMIC-CXR")
metadata = pd.read_csv(root / "train.csv")
image = Image.open(root / "images" / metadata.loc[0, "image"])
Español
Descripción del Dataset
Esta carpeta contiene un subconjunto limpio de MIMIC-CXR preparado para el Medical AI Datathon. Las imágenes son radiografías de tórax preprocesadas en formato JPG.
Los CSV incluyen metadatos del estudio, variables demográficas, etiquetas tipo
CheXpert y el reporte radiológico completo en la columna report.
Dataset original: https://physionet.org/content/mimic-cxr/
Estructura
MIMIC-CXR/
├── images/
├── train.csv
├── valid.csv
├── test.csv
└── README.md
La columna image contiene solo el nombre del archivo. Para cargar una imagen,
combina la ruta del dataset, images/ y el valor de image.
Archivos
images/: radiografías de tórax en formato JPG.train.csv,valid.csv,test.csv: archivos por split con nombres de imagen, metadatos, etiquetas y reportes.README.md: este archivo.
Variables Principales
image: nombre del archivo dentro deimages/.dicom_id,subject_id,study_id: identificadores de imagen, paciente y estudio.split: partición del dataset.ViewPosition: vista radiográfica, por ejemplo AP o PA.Rows,Columns: dimensiones de la imagen.StudyDate,StudyTime: metadatos temporales del estudio.race,sex,age: variables demográficas.race_label,sex_label: variables demográficas codificadas.report: texto completo del reporte radiológico.
Columnas de etiquetas:
AtelectasisCardiomegalyConsolidationEdemaEnlarged CardiomediastinumFractureLung LesionLung OpacityNo FindingPleural EffusionPleural OtherPneumoniaPneumothoraxSupport Devices
Para estas etiquetas, 1 indica presencia, 0 ausencia, -1 incertidumbre y
un valor vacío indica que no hay etiqueta disponible.
Tareas Posibles
- Clasificación multi-etiqueta de radiografías de tórax.
- Modelado multimodal o con reportes usando la columna
report. - Análisis por subgrupos o equidad usando metadatos como edad, sexo y raza.
Ejemplo de Lectura
from pathlib import Path
import pandas as pd
from PIL import Image
root = Path("PATH-TO-DATASET/MIMIC-CXR")
metadata = pd.read_csv(root / "train.csv")
image = Image.open(root / "images" / metadata.loc[0, "image"])
This dataset is a derived subset of the MIMIC-CXR Database.
License & Access Terms
In accordance with the PhysioNet Credentialed Health Data Use Agreement 1.5.0, this data is subject to strict credentialing restrictions:
- PhysioNet Credentialing Required: Access to this subset is strictly limited to researchers who have already signed the MIMIC-CXR Data Use Agreement on PhysioNet and completed the required CITI human subjects training.
- No Redistribution: Users may not redistribute or share this data with third parties or individuals who lack active PhysioNet credentialing.
- No Re-identification: Users strictly agree never to attempt to re-identify any patient or institution.
To request access, click "Request Access" and provide your PhysioNet username/email so access rights can be verified.
- Downloads last month
- 17