Current trends of segmentation algorithms for skin lesions

Skin cancer has become one of the most frequent forms of cancer nowadays;its high prevalence has attracted many studies towards the causes and treatmentsin the recent years. However, the current practice of detecting skin cancersis fairly subjective and may suffer from diagnostic errors. In order to solve thisproblem, an effective computer-aided diagnosis (CAD) system is urgently demanded.Such system can provide an objective source to help the dermatologistimprove the diagnostic accuracy. Such an automated system aims to detect theskin lesions on the acquired images and then analyzes whether those lesions arebenign or malignant. The usual computational procedure is composed of threesteps: image segmentation, feature extraction, and classification. Among thesesteps, the segmentation has deterministic influences to the later quantitative analysisand classification; however, due to the complicated appearance of skin lesionsin the images, correct segmentation of their boundaries is very challenging. Manyalgorithms have been proposed to fulfill this task, and some of them have achievedsatisfactory performances. Nevertheless, the performance of the existing algorithmsstill needs further improvement to be accepted in clinical practice.This paper will review these algorithms and summarize their trends of the development;algorithms focused in this work contain both the ones for dermoscopicimages and the ones for macroscopic images. Advantages and disadvantages ofeach algorithm will be discussed; and possible techniques that can be used for improvementwill be proposed. Open image database will be used for testing and forthe illustration and comparisons among the different algorithms