Published Paper Details:

Skin cancer is a critical global health issue, and early and accurate classification of skin lesions can significantly improve clinical outcomes. Despite the potential of automatic skin cancer classification systems, challenges such as data imbalance, limited training images, cross-domain adaptability, and model robustness persist. Recently, deep learning-based methods have shown promise in addressing these challenges. This paper provides a comprehensive overview of the latest deep learning algorithms for skin cancer classification. We discuss three types of dermatological images and review publicly available skin cancer datasets. Successful applications of typical convolutional neural networks (CNNs) are examined, highlighting solutions for frontier problems like data imbalance, data scarcity, domain adaptation, and model efficiency. Our review indicates that future development in this field is trending towards structured, lightweight, and multimodal approaches. For ease of reference, key findings are summarized in figures and tables.