Lesion border detection in Buruli ulcer images

Buruli ulcer, a disease caused by infection with Mycobacterium ulcerans, is one of the most neglected but treatable tropical diseases. In this paper, a novel segmentation scheme is proposed to detect the border of Buruli lesions in cross-polarization dermoscopic images that were obtained under white light illumination. The method consists of three main steps: first, segmentations in different color spaces by thresholding are fused to form an initial contour, then a level set segmentation is applied to both the luminance and color components of the image, and finally, decisions for each pixel are made by a support vector machine classifier. Experimental results with 26 images show that the proposed methodology outperforms other state-of-art segmentation approaches for Buruli images.