ROCO Instruct 65K: Radiology Images with Captions for Medical AI

Roco Instruct 65K is a dataset of 65,000 radiology images derived from the Radiology Objects in COntext (ROCO) collection. The images were sourced from publications on the PubMed Central Open Access FTP mirror and reformatted for the LLaVA model in the BabyDoctor project. The dataset was created by author photonmz and last updated on August 4, 2023.

Use Cases

• Training vision-language models for radiology image captioning based on the included captions.
• Developing diagnostic AI assistants for medical imaging based on the deep analysis focus mentioned in the description.
• Fine-tuning models for medical concept extraction based on the mentioned UMLS and keywords.
• Creating instructional datasets for medical education based on the structured image-context pairs.

Strengths

• Derived from the large-scale ROCO dataset, a known medical imaging collection.
• Contains 65,000 data points, indicating a substantial size.
• Specifically formatted for the LLaVA model, suggesting a targeted application.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2023-08-04 03:39:56; freshness should be verified.

Provenance

Source

PubMed Central Open Access FTP mirror.

Collection Method

Derived and reformatted from the Radiology Objects in COntext (ROCO) dataset.

Freshness

Last updated 2023-08-04 03:39:56.