A general algorithm for automatic lesion segmentation in dermoscopy images

Abstract

Melanoma is one of the most dangerous types of skin cancer and causes thousands of deaths worldwide each year. Recently dermoscopic imaging systems have been widely used as a diagnostic tool for melanoma detection. The first step in the automatic analysis of dermoscopy images is the lesion segmentation. In this article, a novel method for skin lesion segmentation that could be applied to a variety of images with different properties and deficiencies is proposed. After a multi-step preprocessing phase (hair removal, illumination correction, etc.), a robust histogram-based thresholding technique is used to obtain an initial mask of the lesion. The initial mask is used to sample the lesion's color and drive a contour propagation algorithm. A color probability map of the image is calculated based on sampled pixels and Bayesian classification. Using this probability map and image gradient, a novel dual-component speed function is constructed to improve the performance of propagation model. The proposed algorithm has been tested on the ISIC dataset of 900 dermoscopy images, and gained high values for evaluation metrics e.g. Dice and Jaccard coefficient values of 0.89 and 0.79, respectively. Also, the proposed algorithm ranked 5 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> in the ISBI melanoma segmentation challenge.