Evaluating Public Awareness of Skin Cancer Risk Factors and Adherence to Sun-Safety Behaviors

Abstract

Skin cancer is the most common cancer in the world and is the most common form of cancer reported in the world each year which correlates to a rise in exposure to ultraviolet (UV) radiation and ozone depletion. Although the field of developing deep learning-based diagnostic systems is advancing rapidly, public understanding of the risk factors for skin cancer and practices of sun protection behaviors such as use of sun protection products and sun protection clothing are still lacking. These preventive practices are different in various populations that make it difficult to do real-world intervention. In this research, a hybrid deep learning architecture for multi-class skin lesion classification networks is proposed, Sun Awareness Intelligent Diagnosis Network (SAID-Net), which can explain the diagnosis, to solve both these problems. It incorporates EfficientNetV2, a local feature extractor for texture features, Vision Transformer with Multi-Head Self-Attention (ViT-MHSA), which is a global feature extractor for contextual information and Explainable AI (XAI) Attention module, which identifies and highlights clinically relevant lesion areas for greater transparency during diagnosis. Pre-processing consists of dermoscopic hair removal, image contrast enhancement by CLAHE and normalization of images. Evaluated on the ISIC dataset, SAID-Net achieves 99.12% accuracy, outperforming CNN (92.84%), ResNet50 (94.27%), ASFF (95.36%), LSTM (96.14%) and Denoising Autoencoder (97.03%). The integrated system can be used in detecting skin cancer automatically and can be applied in public campaigns for sun safety.