Skin disease segmentation: current practices, techniques and emerging trends

Skin disease segmentation is of paramount importance when it comes to early detection and implementation of diagnosis for skin related diseases for the management of proper health interventions. This review aims at stating the current situation in skin disease segmentation, both in terms of existing methodologies and new trends that have emerged in recent years, with an emphasis on the combination of standard image processing methods and state-of-art machine learning and deep learning techniques. The paper discusses the strength and weakness of the above techniques and how annotated datasets, preprocessing techniques, as well as proper metrics contribute to the performance of segmentation. Furthermore, it examines factors like skin tone and lesion variability, the variety of lesions which are difficult to depict, and lighting conditions that may affect the performance of the model. Recorded trends such as multimodal data fusion, transfer learning and explainable artificial intelligence are also discussed as potential directions towards the above challenges. The present review seeks to give a brief comprehensive overview of the dynamic state of skin disease segmentation, and the findings may be useful in facilitating further research and development into this important area.