Deep Learning for Skin Lesion Segmentation: A Review and Outlook

Human skin acts as an important barrier that protects the proper functioning of the body's circulatory system. According to surveys in the last 30 years, the number of people affected by skin lesions far exceeds that by all other cancers combined. Currently, skin lesion is mainly diagnosed by microscopic or dermoscopic analysis, which heavily relies on the experience of the pathologist. It is highly subjective and time-consuming. Considering these concerns, the computer-aided diagnosis (CAD) approach has been proposed for skin lesion analysis. It is beneficial for doctors to make surgical plans and improve the prognosis of patients. This article aims to provide a comprehensive survey and review of recent advances in an area of increased interest for skin lesion segmentation used deep learning methods between 2020 and 2023. This survey includes relevant and essential definitions and theories, datasets utilization, deep learning methods, some techniques for skin lesion segmentation and evaluation metrics. In addition, we discuss the main challenges encountered for medical image segmentation of skin lesions. Finally, some important yet under-investigated issues are discussed that help researchers contribute to future research.