10,015 dermatoscopic images of pigmented skin lesions across seven diagnostic categories. The dataset provides ground truth labels for both multi-class classification and pixel-level lesion segmentation tasks.
Use Cases
- Train a multi-class classifier to identify skin pathologies using the dx label and image files
- Build a segmentation model to isolate lesions from surrounding skin using the provided mask images
- Analyze the correlation between lesion types and anatomical sites using the localization column
- Investigate demographic trends in skin cancer types using the age and sex metadata
Strengths
- 10,015 high-resolution dermatoscopic images covering seven diagnostic classes including melanoma and basal cell carcinoma
- Includes metadata for every image such as patient age, sex, and anatomical localization
- Ground truth segmentation masks provided for training boundary detection algorithms
- Diagnostic labels (dx) verified through histopathology, follow-up, or expert consensus