Transforming Skin Cancer Diagnosis: A Novel Approach Using Vision Transformer Networks

Skin cancer is one of the most common kinds of cancer, and its early detection is crucial for successful treatment. Automated skin cancer diagnosis has recently been made possible due to a variety of computer-aided techniques. These techniques entail examining digital photographs of skin lesions to find probable cancers. The accuracy of skin cancer detection models has greatly increased because of the advent of deep learning techniques. Convolutional neural networks (CNNs) have demonstrated results nearly similar to those attained from dermatologist diagnosis results while identifying and classifying skin cancer. In conclusion, machine learning and deep learning approaches for computer-aided skin cancer diagnosis have enormous potential to enhance the early identification of skin cancer and thus reduce mortality rates. But the deep learning and machine learning models performance is often hindered by inconsistent annotations and imbalances between classes. We proposed a novel hybrid model LeViT which is a combination of ConvNets and Vision Transformer architectures that improved the classification accuracy of skin cancer at early stages. Our proposed model efficiency is compared with other existing state of art architectures employed for classifying skin cancer.