Application of Artificial Intelligence in Skin Cancer Diagnosis: Convolutional Neural Network-Based Methods

Abstract

Skin cancer, a common malignant tumor, poses a significant global health burden, with traditional diagnosis methods relying primarily on visual observation, which often lack accuracy and consistency. In recent years, Artificial Intelligence (AI), particularly deep learning, has made notable advancements in medical imaging diagnosis, exemplified by systems such as Google’s LYNA and various AI-based skin cancer classification models. This paper reviews skin cancer diagnosis using Convolutional Neural Networks (CNNs) in recent two years, covering baseline CNNs e.g., VGG16, MobileNet, attention-based CNNs e.g., Graph Attention Network (GAT)+SENet, Convolutional Block Attention Module (CBAM), and hybrid models e.g., CNN-Transformer, Convolutional Neural Network-Recurrent Neural Network (CNN-RNN). These methods enhance accuracy via transfer learning, attention mechanisms, and hybrid architectures. Challenges include model interpretability (black-box nature), generalization across diverse populations, and data privacy. Future directions point toward the development of white-box models, standardized evaluation benchmarks for generalization, and the adoption of federated learning frameworks to enable privacy-preserving collaborative model training. This review aims to provide a comprehensive overview to advance the field.