A system for the detection of melanomas in dermoscopy images using shape and symmetry features

摘要

Different Computer Aided Diagnosis (CAD) systems have been proposed to diagnose skin lesions in dermoscopy images. nMost of them use different types of features inspired in the ABCD rule: asymmetry, border, color and differential structures. However, it is not clear which are the most relevant types of features. In this paper we present a study on the importance of shape and symmetry features in the detection of melanomas. In order to assess the relevance of these features, a CAD system that solely uses symmetry or shape features to represent lesions was developed. In this manner, we are able to evaluate the contribution of each one of these types of features. The system was tested using the PH2 database of annotated images from the Hospital Pedro Hispano. The CAD system achieved a sensitivity of 83%, specificity of 78% with shape features, and a sensitivity of 96% and specificity of 86%, using symmetry features. We conclude that both types of features convey discriminative information about the lesion type but symmetry plays the most important role. A comparison with color and texture features is also provided.