SynthVision Medical Image Seeds from Four Open Datasets

119,137 seed records aggregated from four open medical image datasets, including ROCO-radiology, multicare-images, path-vqa, and vqa-rad. The dataset, created by OpenMed, is deduplicated by SHA-256 hash and serves as a foundation for the SynthVision pipeline.

Use Cases

• Train a multi-modal classifier using records labeled with Radiology, Pathology, and Mixed modalities.
• Develop a vision-language model using seed records from path-vqa and vqa-rad datasets, which are associated with visual question answering tasks.
• Perform deduplication analysis on medical images using the SHA-256 hash field to ensure dataset uniqueness.
• Analyze the distribution of records across the four source datasets (ROCO-radiology, multicare-images, path-vqa, vqa-rad) to understand composition.

Strengths

• Contains 119,137 records, providing a substantial base for medical image analysis.
• Aggregates data from four distinct open datasets, offering diversity in medical imaging modalities.
• Records are deduplicated by SHA-256 hash, ensuring uniqueness and reducing redundancy.

Limitations

• Specific column definitions, file formats, and image sizes are not detailed in the provided input.
• The sample data is unavailable, preventing immediate inspection of record structure and content.
• The dataset's license, geographic coverage, and temporal range are unspecified.

Provenance

Source

Aggregated from four open datasets: eltorio/ROCO-radiology, OpenMed/multicare-images, flaviagiammarino/path-vqa, and flaviagiammarino/vqa-rad.

Collection Method

Records were aggregated and deduplicated by SHA-256 hash to create seed data for the SynthVision pipeline.

Time Range

null

Freshness

null

Geography

null