TorchXRayVision: Unified Access to Chest X-Ray Datasets and Models

TorchXRayVision is a collection of chest X-ray datasets and pre-trained models developed by mlmed and updated through September 2025. It provides a unified programmatic interface for accessing multiple medical imaging sources to support classification, segmentation, and autoencoding tasks.

Use Cases

• Multi-label classification of pathologies such as pneumonia or cardiomegaly using standardized diagnostic labels
• Anatomical segmentation of lungs and heart using provided segmentation tools and masks
• Transfer learning using pre-trained weights from chest X-ray foundation models for specialized medical AI tasks

Strengths

• Unified API for accessing multiple major chest X-ray datasets like ChestX-ray14 and PadChest
• Includes pre-trained weights for classification and segmentation tasks
• Apache-2.0 licensed for open research and development

Limitations

• Requires separate data access agreements and manual downloads for restricted source datasets
• Label accuracy is dependent on the automated extraction methods used by the original data providers

Provenance

Source

mlmed (GitHub)

Collection Method

Aggregated from multiple primary medical imaging repositories and research institutions

Freshness

Updated as of September 2025.