Deep Skin Cancer Lesions Classification Scheme

Skin cancer has risen to be one of the significant forms of cancer worldwide. However, the traditional means of skin cancer detection requires manual examination by an expert, which is often time-consuming. Recently, computer-aided methods have garnered much traction in the research community and among medical practitioners. This paper proposes using a deep-learning model to classify skin lesions into seven classes. The model is a fine-tuned 16-patch vision transformer (ViT) model and a custom classifier. The custom classifier consists of a convolutional computation fitted to the end of the ViT model. Experimental results show that the proposed scheme can classify various (7) skin lesions with an accuracy of 98%.