Skin cancer, especially melanoma, is the most common cancer globally and
in the United States. About 10% of melanoma cases have a family history. In 2023, an
estimated 97,610 new melanoma cases are expected in the US, with 58,120 affecting
males and 39,490 affecting females. To enhance diagnostic accuracy, the DeiT (DataEfficient Image Transformer) model was applied to the ISIC 2019 dataset, which
comprises a diverse range of skin lesion images. Our Transformer model outperformed
traditional CNN models, achieving a precision of 94%, a recall of 95%, and an
accuracy of 93%, surpassing models such as EfficientNet and DenseNet169. This
superior performance is attributed to the DeiT model's ability to capture complex image
patterns efficiently. Its success indicates that Transformer-based models hold
significant promise for improving skin cancer diagnosis, leading to earlier detection
and better patient outcomes.
Keywords: DeiT, Melanoma, Skin Cancer, United States, Worldwide.
