Highly Accurate Skin Cancer Diagnosis Using HMT-NET and Vision Transformer Models

Skin cancer, especially melanoma, is one of the most aggressive and fatal cancers, with rising global incidence driven largely by ultraviolet exposure. Early detection is critical, as visual similarities among lesions complicate diagnosis. Clinical methods include self-examination, dermoscopic, and biopsy, while computational approaches range from traditional Computer-aided design systems to deep learning models like convolutional neural network and Vision Transformers. Although these models enhance accuracy, they face limitations such as high data requirements, limited interpretability, and inconsistent image quality, emphasizing the need for scalable and explainable diagnostic systems in clinical settings. A Hybrid Multi-layer Transformer Network (HMT-NET) and Vision Transformer (ViT)-based model is proposed for accurate skin lesion segmentation and multiclass skin cancer classification. This study presents a hybrid HMT-NET and ViT framework for accurate skin lesion segmentation and classification. Evaluated on HAM10000 and ISIC2019 datasets, the model achieved high Dice score and Jaccard index and 98.75% classification accuracy, demonstrating superior performance and reliability for integration into automated dermatological diagnostic systems.