PAD_UFES_20: Smartphone-Captured Skin Lesion Images for Diagnostic Classification

Smartphone-captured images of skin lesions for diagnostic classification into six categories. The dataset includes 16 reasoning traces per example, generated by GPT-4o, available in a responses column. It was created by OctoMed and is hosted on Hugging Face, with a last recorded update in April 2026.

Use Cases

• Train skin lesion classification models based on smartphone-captured images.
• Develop or evaluate explainable AI (XAI) methods based on the provided reasoning traces.
• Benchmark model performance on a multi-class diagnostic task for dermatology.
• Study the utility of large language model-generated annotations for medical image analysis.

Strengths

• Includes 16 reasoning traces per image example, generated by GPT-4o.
• Focuses on a specific diagnostic task with six classification categories.
• Builds from an original, cited data source hosted on Mendeley.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the original collection source and smartphone capture methods.

Provenance

Source

Original data source available at https://data.mendeley.com/datasets/zr7vgbcyr2…

Collection Method

Smartphone-captured images, with reasoning traces sampled using GPT-4o.

Freshness

Last updated 2026-04-28 03:22:36; freshness should be verified.