NIH Chest X-ray: 100,000 De-identified Images for Pathology Classification

The NIH Clinical Center released this collection of 100,000 de-identified chest X-ray images in PNG format for medical research. The data supports multi-class image classification tasks and is documented in the research paper Arxiv 1705.02315.

Use Cases

• Multi-class classification of thoracic pathologies using PNG image data
• Benchmarking diagnostic AI models against the Arxiv 1705.02315 baseline
• Research into automated radiology report generation using image-label pairs

Strengths

• 100,000 clinical images
• Expert-generated annotations
• Standardized PNG format

Limitations

• Label noise may exist due to machine-generated annotations derived from radiology reports
• Geographic bias as data is sourced from a single US-based clinical center

Provenance

Source

NIH Clinical Center

Collection Method

De-identified clinical records and radiology reports

Freshness

Last updated September 2024; based on a 2017 research publication.

Geography

United States