Early Melanoma Detection Based on Chromatic Descriptors and Machine Learning Algorithms

Abstract

This article proposes a pigmented nevi classification methodology such to early detect melanoma. This is a deadly type of skin cancer with continuous growth in incidence. The strategy that we propose is based on chromatic descriptors of the skin lesions and uses support vector machines for classification. On a 604 images database (138 of melanomas), we obtain a sensitivity and specificity of 78% and, respectively, 82%. Out of 19 images of melanoma in situ, only one was misclassified as benign.