On the role of shape in the detection of melanomas

This paper presents a study on the role of shape in the detection of melanomas in dermoscopy images. The contribution of shape-related features was assessed by developing a Computer-Aided Diagnosis (CAD) system whose classification is solely based on this type of features. Four shape descriptors were used, first separately and then simultaneously, to describe the images. Image segmentation was performed both manually, by an expert, and automatically, by using an Adaptive Thresholding algorithm. The best performances were SE = 92% and SP = 74%, obtained for manually segmented images, and SE = 92% and SP = 78%, obtained for automatically segmented images. These results were achieved by combining the shape descriptors and show that shape information plays an important role in melanoma detection. Furthermore, no degradation was observed when automatic segmentation methods are used, instead of manual ones.