81,000 radiology images paired with natural language captions extracted from PubMed Central. The dataset includes diverse modalities such as X-ray, CT, MRI, and Ultrasound, categorized into radiology and non-radiology classes.
Use Cases
- Train image captioning models using the image and caption pairs to generate descriptive medical text
- Develop cross-modal retrieval systems to find relevant radiology images based on text queries from the caption field
- Build a classification model to filter non-radiology images using the provided category labels
Strengths
- 81,000 image-caption pairs sourced from PubMed Central
- Includes binary classification labels to distinguish radiology images from non-radiology content like synthetic data or clinical photos
- Covers multiple imaging modalities such as CT, MRI, Ultrasound, and X-ray
- Structured into training, validation, and test splits for standardized benchmarking