High-quality annotated histopathological skin melanoma subtype slides for AI model development and validation

Abstract

Cutaneous melanoma (CM) is a highly aggressive form of skin cancer requiring accurate subtype classification for effective prognosis and treatment. To address the scarcity of high-quality datasets, we curated 82 Hematoxylin & Eosin (H&E)-stained whole slide images (WSIs) from 32 patients in the Clinical Proteomic Tumor Analysis Consortium (CPTAC)-CM collection. A senior dermatopathologist oversaw a rigorous multi-stage annotation process, involving a medical trainee and a board-certified pathologist, aligning with the 2022 World Health Organization (WHO) classification. The curation process resulted in a robust dataset comprising 196 tissue sections annotated with histopathologic biomarkers and prognostic metadata, such as ulceration status and tumor cell proportion (TCP). Furthermore, statistical validation using Chi-squared and Fisher’s exact tests confirmed that our final selection represents the original cohort without demographic or histological sampling bias, evidenced by a p-value greater than 0.05. Comprising 196 tissue sections in high-resolution SVS format, this expert-annotated dataset serves as a foundation for developing deep learning AI models to enhance dermatopathology and personalized patient care.