Divergence-based colour features for melanoma detection

Abstract

Melanoma is a deadly form of skin cancer which is difficult to detect in its early stages. Several computer-aided diagnostic systems based on dermoscopic images of skin lesions intend to improve melanoma detection. Colour is an important factor in correctly classifying a skin lesion. Here, we introduce divergence-based colour features, using the Kullback-Leibler information as a preferred divergence function. These features are based on the divergence between the distribution of the pixel values of a lesion image, and that of the pixel values of either a benign or a malignant model. The features’ sensitivities and specificities are reported, along with the contribution to an existing classifier for skin lesions. The features improve the performance of the existing classifier and are therefore relevant for melanoma detection.