A new risk assessment methodology for dermoscopic skin lesion images

Abstract

The incidence of melanoma has been increasing steadily over the past few decades throughout most of the world. The development of computer diagnosis systems that use dermoscopic images can be of great help for the diagnosis of melanoma. This paper presents an image processing and analysis methodology using supervised classification to independently assess the Asymmetry, Border, Color and Dermoscopic Structures score according to the ABCD rule, and the corresponding Total Dermatoscopy Score of a skin lesion using dermoscopic images. A dermoscopic image dataset was used to test the proposed approach, annotated by dermatology specialists according to the ABCD rule and being the confirmed malignant melanomas also identified. Accuracy rates of 74.0%, 78.3% and 53.5% were achieved for the estimation of the ABCD score of the Asymmetry, Border and Color criterion, as well as accuracy rates for the presence of the five Differential Structures of 72.4%, 68.5%, 74.0%, 74.0% and 85.8% for dots, globules, streaks homogeneous areas and pigment network. Moreover, sensitivity and specificity rates of 93.3% and 69.1% were achieved for the classification of the dermoscopic images as melanoma or non-melanoma.