ReXGradient-160K: 273,004 Chest X-Rays and Reports from 79 U.S. Sites

ReXGradient-160K contains 273,004 unique chest X-ray images from 160,000 radiological studies across 79 medical sites in the United States. Developed by the Rajpurkar Lab and updated in late 2025, the dataset links medical imagery with detailed radiology reports for 109,487 unique patients.

Use Cases

• Training vision-language models using paired chest X-ray images and radiology reports
• Automated generation of clinical findings from medical imagery
• Evaluating AI model generalization across 79 distinct medical environments

Strengths

• 273,004 unique chest X-ray images
• Data sourced from 79 medical sites across 3 U.S. health systems
• Includes paired radiology reports for 160,000 studies
• Large patient cohort of 109,487 individuals

Limitations

• Geographic bias as data is limited to the United States
• Potential variability in report formatting and terminology across 79 different sites
• Specific license restrictions under 'Licenseother' category

Provenance

Source

Rajpurkar Lab, Arxiv 2509.13270 and 2505.00228

Collection Method

Aggregated from 3 U.S. health systems

Freshness

Last updated December 2025; based on 2025 research publications.

Geography

United States