Revolutionizing Skin Cancer Detection Merging CNNs with Vision Transformers

Skin cancer is one of the top health concerns globally, and early detection is essential to ensure good patient outcomes. Earlier research studies have applied CNNs or Vision Transformers separately, and CNNs outperform on local feature extraction, whereas ViTs are more successful in capturing global dependencies. However, such isolated techniques suffer from limitations such as lower accuracy on heterogeneous datasets and poor generalization. This work bridges this gap by introducing a hybrid model that synergizes the strengths of both architectures. The CNN extracts low-level details like textures and edges, while the ViT captures spatial relationships across the image. Evaluated on open-source skin cancer datasets, the proposed model outperforms individual approaches in accuracy, reliability, and generalization, making it a robust tool to assist dermatologists in early stage diagnosis.