Skin_Cancer_PAD-UFES-20: 2,298 Skin Lesion Images with 26 Clinical Features

PAD-UFES-20 is a dataset of 2,298 skin lesion images from 1,373 patients, published in July 2020 by researchers from the Federal University of Espirito Santo, Brazil. It contains six types of lesions, including three skin cancers, with approximately 58% of samples being biopsy-proven. Each image is associated with up to 26 clinical metadata features such as patient demographics, lesion characteristics, and family history.

Use Cases

• Classifying skin lesion types from images based on the six diagnostic categories.
• Predicting biopsy-proven status using clinical metadata features like age, skin type, and family history.
• Analyzing correlations between lifestyle factors (smoking, alcohol, pesticide exposure) and lesion diagnosis.
• Developing multimodal models that combine image data with tabular clinical features for improved diagnostic accuracy.

Strengths

• Contains 2,298 images and 26 metadata features per instance.
• Approximately 58% of samples have biopsy-proven ground truth.
• Includes data from 1,373 patients, providing multiple lesions per patient for some cases.

Limitations

• Image sizes vary due to collection from different smartphone devices, which may require preprocessing.
• The last update date is unknown, so dataset freshness is unverified.
• Not all samples are biopsy-proven; some diagnoses are based on clinical consensus.

Provenance

Source

Pacheco, A. G., Lima, G. R., Salomao, A. S., Krohling, B., Biral, I. P., de Angelo, G. G., ... & de Barros, L. F. from the Federal University of Espirito Santo, Brazil.

Collection Method

Images collected using different smartphone devices; metadata compiled from patient records and clinical diagnosis.

Time Range

Date of publication is July 2020; specific collection period is not stated.

Freshness

Last updated date is unknown.

Geography

Likely Brazil, given the author's institution, but not explicitly stated.