You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

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 inside images/.
  • 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:

  • Atelectasis
  • Cardiomegaly
  • Consolidation
  • Edema
  • Enlarged Cardiomediastinum
  • Fracture
  • Lung Lesion
  • Lung Opacity
  • No Finding
  • Pleural Effusion
  • Pleural Other
  • Pneumonia
  • Pneumothorax
  • Support 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 report column.
  • 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 de images/.
  • 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:

  • Atelectasis
  • Cardiomegaly
  • Consolidation
  • Edema
  • Enlarged Cardiomediastinum
  • Fracture
  • Lung Lesion
  • Lung Opacity
  • No Finding
  • Pleural Effusion
  • Pleural Other
  • Pneumonia
  • Pneumothorax
  • Support 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

Collection including dsrestrepo/mimic-cxr-datathon