Classification of Skin Cancer Using Integrated Methodology

Skin diseases are prevalent in human life and can range from mild to severe, causing significant deterioration in individuals. These diseases come in various types, characterized by color and infection, where some issues may be easily overlooked while others can lead to cancer. Our project proposes an approach for automatically diagnosing malignant or benign conditions. Despite numerous attempts by researchers using deep learning models, accurately analyzing small patches of skin images has remained challenging. To address this problem, we employed a vision transformer model trained on small sequential patches of skin images. Specifically, we utilized the Skin Cancer Detection Dataset (ISIC), dividing each image into a grid of 16×16 patches and training a sequence model accordingly. This enabled our model to diagnose small patches within the image effectively. Compared to other established models, our approach demonstrated superior performance, as evident from precision and recall values of 0.87 and 0.89, respectively.