Skin Disease Images with Medical Descriptions

SkinCAP is a multi-modal dataset of 4,000 dermatology images annotated with medical descriptions. The dataset was created by author joshuachou and last updated in July 2024. Images were sourced from the Fitzpatrick 17k and Diverse Dermatology Images datasets.

Use Cases

• Train image-to-text models to generate medical captions from skin disease images.
• Fine-tune vision encoders for skin disease classification using the annotated image labels.
• Develop multi-modal retrieval systems to find similar cases based on image features and caption text.
• Benchmark the performance of large language models on generating or understanding dermatological descriptions.

Strengths

• 4,000 annotated dermatology images.
• Annotations include medical descriptions created by dermatologists.
• Data sourced from two established dermatology image collections.

Limitations

• Sample size of 4,000 images may be limited for training large models from scratch.
• Annotations were machine-translated to English, which may introduce errors.
• Geographic and demographic coverage of source datasets may be biased.

Provenance

Source

Fitzpatrick 17k skin disease dataset and Diverse Dermatology Images dataset.

Collection Method

Images sourced from existing datasets and annotated by dermatologists.

Time Range

null

Freshness

Last updated July 2024.

Geography

Region tag indicates 'us', suggesting United States coverage.