Intelligent Dermatology: Exploration of Transformer Networks for Skin Disease

Abstract – The objective of this paper is to explore various existing Transformer networks for diagnostic and classification ofcomplex patterns in dermatological images. Initially designed for natural language processing, transformer networks have shown impressive effectiveness in computer vision tasks, including the detection of skin diseases. In contrast to convolutional neural networks (CNNs) that analyze localized image regions, Transformers process the entire image at once, enabling the capture of both intricate details and broader contextual patterns. This survey reviews how different Transformer models contribute to skin disease classification by enhancing feature extraction, offering better scalability, and achieving improved performance on benchmark datasets such as the International Skin Imaging Collaboration (ISIC), the ISIC Challenge, and HAM10000. The findings emphasize the potential of Transformers to increase diagnostic accuracy, mitigate challenges related to limited data, and support the development of more efficient and interpretable diagnostic systems.