A Review of U‐Net‐Based Deep Learning Models for Skin Lesion Segmentation

ABSTRACT Automated skin lesion segmentation is crucial for early and accurate skin cancer diagnosis. Deep learning, particularly U‐Net, has revolutionized the field of automatic skin lesion segmentation. This review comprehensively examines U‐Net and its variants employed for automated skin lesion segmentation. It outlines the foundational U‐Net architecture and explores diverse architectural innovations, including attention mechanisms, advanced skip connections, residual and dilated convolutions, transformer models, and hybrid models. The review highlights how these adaptations address inherent challenges in skin lesion segmentation, including data limitations and lesion heterogeneity. It also discusses the commonly used datasets, evaluation metrics, and compares model performance and computational cost. Finally, it addresses the existing challenges and outlines future research directions to advance automated skin cancer diagnosis.