DERM12345: A Large, Multisource Dermatoscopic Skin Lesion Dataset with 40 Subclasses.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin lesion datasets are crucial for developing diagnostic tools and advancing AI in dermatology.
• Existing datasets often lack comprehensive coverage of skin lesion subclassifications, limiting AI model performance.
• There is a need for larger, more diverse datasets to improve the accuracy of skin lesion analysis and reduce prediction errors.

Purpose of the Study:

• To present a novel, diverse dermatoscopic image dataset for skin lesion research.
• To provide a reliable foundation for AI-driven skin cancer detection and analysis.
• To facilitate targeted research and enhance understanding of various skin lesions.

Main Methods:

• Compilation of 12,345 high-resolution dermatoscopic images from Turkiye.
• Inclusion of expert annotations for each image.
• Dataset structured into 5 super classes, 15 main classes, and 40 subclasses of skin lesions.

Main Results:

• A comprehensive dataset of 12,345 dermatoscopic images representing 40 skin lesion subclasses.
• The dataset captures diverse skin types from Turkiye, a region bridging Europe and Asia.
• Expert annotations provide a reliable basis for machine learning model training and validation.

Conclusions:

• The presented dataset offers a valuable resource for advancing AI in dermatology.
• Its diversity and detailed structure are expected to improve the accuracy of skin lesion classification and reduce diagnostic errors.
• This dataset will support future research in early skin cancer detection and treatment planning.