HAM10000: 10,015 Multi-Source Dermatoscopic Images of Pigmented Skin Lesions

10,015 dermatoscopic images were collected to address the small size and lack of diversity in datasets for training neural networks to diagnose pigmented skin lesions. The dataset includes a representative collection of seven important diagnostic categories, with more than 50% of lesions confirmed through histopathology. It was released by Philipp Tschandl of the Medical University of Vienna and served as the training set for the ISIC 2018 challenge.

Use Cases

• Training neural networks for automated diagnosis of pigmented skin lesions based on multi-source dermatoscopic images.
• Benchmarking model performance on a standardized dataset used for the ISIC 2018 challenge.
• Studying the classification of seven common diagnostic categories, including melanoma and basal cell carcinoma.
• Analyzing cases with multiple images per lesion, trackable via a lesion_id column.

Strengths

• Contains 10,015 dermatoscopic images, a large collection for academic machine learning purposes.
• Covers seven important diagnostic categories of pigmented lesions, providing representative diversity.
• More than 50% of lesions have ground truth confirmed by histopathology, with the rest confirmed by follow-up, expert consensus, or confocal microscopy.
• Includes an additional test set of 1,511 images from the ISIC 2018 challenge for evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data may reflect geographic or source bias inherent to the contributing populations and modalities.

Provenance

Source

Medical University of Vienna

Collection Method

Collected dermatoscopic images from different populations, acquired and stored by different modalities.