Study on different Skin lesion Segmentation techniques and their comparisons

Abstract

Skin cancer is increasing rapidly all around the world with an enhanced mortality rate every year. Melanoma is one of the most frequent skin cancer diseases which is curable if diagnosed at an early stage. In the present scenario, one of the challenging tasks for a dermatologist is to precisely detect and classify skin lesion dermoscopic images. These images have complex structures with various other artifacts. The dedicated skin lesion segmentation technique is a need of an hour to enhance the diagnostic capability. This study focuses on three major steps involved for efficient detection of melanoma skin cancer including preprocessing, segmentation, and postprocessing. K-means clustering, Global thresholding using Otsu’s method, and Chan-Vese Active contour model are used as preparatory steps for skin lesion segmentation. The results are been presented in form of accuracy, dice coefficient, and Jaccard index. The average value of accuracy, dice coefficient, and Jaccard index of 50 dermoscopic skin lesion images are 96.51%, 92.14%, and 86.75% respectively. The comparison of all three techniques is discussed on basis of performance indices. The results are quite convincing and proclaim that the proposed technique could be used for skin lesion segmentation.