Transformers in Skin Lesion Classification and Diagnosis: A Systematic Review

Skin lesion classification is a critical task in dermatology, aiding in the early diagnosis and treatment of skin cancer. In recent years, transformer-based models, originally developed for Natural Language Processing (NLP) tasks, have shown promising results in many classification tasks specifically the image classification domains. This systematic review aims to provide a comprehensive overview of the current state of research on the application of transformers in skin lesion classification. Over the period 2017-2023, this systematic review investigated the application of transformer-based models in skin lesion classification, focusing on 57 articles retrieved from prominent databases which are PubMed, Scopus, and Medline. The inclusion criteria encompass studies centering on transformer-based models for skin lesion classification, utilization of diverse datasets (dermoscopic images, clinical images, or histopathological images), publication in peer-reviewed journals or conferences, and availability in English. Conversely, exclusion criteria filter out studies not directly related to skin lesion classification, research applying algorithms other than transformer-based models, non-academic articles lacking empirical data, papers without full-text access, and those not in English. Our findings underscore the adaptability of transformers to diverse skin lesion datasets, the utilization of pre-trained models, and the integration of various mechanisms to enhance feature extraction.