Skin cancer remains one of the most common malignancies worldwide, with an increasing incidence in recent decades. Early detection is essential to reduce mortality, but traditional diagnostic methods, including visual examination and dermoscopy, are time-consuming and susceptible to inter-observer variability. Artificial intelligence (AI), particularly deep learning, has shown great potential in medical image analysis and skin tumor identification, reaching the level of dermatologists. This review systematically analyzes recent advances in AI in image recognition for skin tumors, with a particular focus on the equity and performance of these models among different skin types and ethnic groups. Through a literature review of studies published in recent years, this study examined public datasets, algorithmic approaches, equity challenges, and clinical integration prospects. The review found that while AI systems show robust diagnostic performance, there are still differences in model accuracy across skin tones due to imbalanced datasets. We conclude that enhancing dataset diversity, improving interpretability, and establishing clinical validation pipelines are essential for achieving equitable AI in dermatology.
Yu-Ching Chiu (Thu,) studied this question.