20 March 2015 Automatic detection and segmentation of vascular structures in dermoscopy images using a novel vesselness measure based on pixel redness and tubularness
Author Affiliations +
Abstract
Vascular structures are one of the most important features in the diagnosis and assessment of skin disorders. The presence and clinical appearance of vascular structures in skin lesions is a discriminating factor among different skin diseases. In this paper, we address the problem of segmentation of vascular patterns in dermoscopy images. Our proposed method is composed of three parts. First, based on biological properties of human skin, we decompose the skin to melanin and hemoglobin component using independent component analysis of skin color images. The relative quantities and pure color densities of each component were then estimated. Subsequently, we obtain three reference vectors of the mean RGB values for normal skin, pigmented skin and blood vessels from the hemoglobin component by averaging over 100000 pixels of each group outlined by an expert. Based on the Euclidean distance thresholding, we generate a mask image that extracts the red regions of the skin. Finally, Frangi measure was applied to the extracted red areas to segment the tubular structures. Finally, Otsu’s thresholding was applied to segment the vascular structures and get a binary vessel mask image. The algorithm was implemented on a set of 50 dermoscopy images. In order to evaluate the performance of our method, we have artificially extended some of the existing vessels in our dermoscopy data set and evaluated the performance of the algorithm to segment the newly added vessel pixels. A sensitivity of 95% and specificity of 87% were achieved.
© (2015) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Pegah Kharazmi, Harvey Lui, William V. Stoecker, Tim Lee, "Automatic detection and segmentation of vascular structures in dermoscopy images using a novel vesselness measure based on pixel redness and tubularness", Proc. SPIE 9414, Medical Imaging 2015: Computer-Aided Diagnosis, 94143M (20 March 2015); doi: 10.1117/12.2082720; https://doi.org/10.1117/12.2082720
PROCEEDINGS
6 PAGES


SHARE
Back to Top