Deep Learning for Classification and Prognosis of Melanoma in Whole-Slide Images: A Review

ABSTRACT: The rising incidence of melanoma highlights the limitations of traditional diagnostic methods, including inefficiency, subjectivity, and poor quantifiability. Consequently, deep learning (DL)-based diagnosis and prognosis using whole-slide images (WSIs) has emerged as a major research focus. This systematic review adhered to PRISMA guidelines to outline the current landscape and trends in this rapidly evolving field, analyzing relevant studies published between January 2020 and June 2024, retrieved from PubMed and Web of Science. After screening, 27 studies were included for qualitative synthesis. The review first details the application of DL in the diagnostic classification of melanoma WSIs, encompassing both binary and multiclass categorization. It further examines the emerging role of DL in prognostic assessment based on WSIs. Through a critical analysis of the included literature, key challenges are identified, primarily concerning the generalizability, interpretability, and clinical integration of these models. Finally, the article proposes targeted future research directions to address these barriers, aiming to guide the translation of DL tools from research to clinical practice and toward more precise management of patients with melanoma.