Skin Cancer Detection and Classification: A Review of Recent Advances

Abstract

Skin cancer is most common diseases find in humans worldwide and early detection can be assist to improve the doctors’ diagnosis. Traditional diagnosis depends on visual examination by dermatologists using dermoscopy also time consuming and depends heavily on the experience of the clinician or doctor. In recent years, deep learning has become the leading approach for automatic detection and classification skin lesions from dermoscopic and clinical images. In this paper we present a focused review of the most recent deep learning methods, published recently, for skin cancer detection and classification. The review covers benchmark datasets, image preprocessing and classification and segmentation methods, and major deep learning architectures, which include convolutional neural networks (CNNs), hybrid CNN-Transformer models, vision transformers, generative adversarial networks (GANs), and multimodal fusion approaches. It also summarizes the main challenges that still limit the real-world use of these systems, such as complex lesion features, image noise, intra-class variation, inter-class similarity, and class imbalance. In this paper we discuss future research directions, such as explainable artificial intelligence, fairness across different skin tones, and uncertainty-aware models. Keywords— Skin cancer; deep learning; dermoscopy; convolutional neural network; vision transformer; image classification; class imbalance