Please cite this article in press as: Mete M, et al. Fast density-based lesion detection in dermoscopy images. Comput Med Imaging Graph (2010), doi:10.1016/j.compmedimag.2010.07.007 ARTICLE IN PRESS G Model CMIG-1020; No. of Pages 9 Computerized Medical Imaging and Graphics xxx (2010) xxx–xxx Contents lists available at ScienceDirect Computerized Medical Imaging and Graphics journal homepage: www.elsevier.com/locate/compmedimag Fast density-based lesion detection in dermoscopy images Mutlu Mete a,∗ , Sinan Kockara b , Kemal Aydin c a Department of Computer Science, Texas A&M University-Commerce, United States b Department of Computer Science, University of Central Arkansas, United States c Department of Computer Science, University of Arkansas at Pine Bluff, United States article info Article history: Received 28 October 2009 Accepted 30 July 2010 Keywords: Dermoscopy Density-based clustering Image understanding CAD abstract Dermoscopy is one of the major imaging modalities used in the diagnosis of melanoma and other pig- mented skin lesions. Automated assessment tools for dermoscopy images have become an important research field mainly because of inter- and intra-observer variations in human interpretation. One of the most important steps in dermoscopy image analysis is automated detection of lesion borders. In this study, we introduce a border-driven density-based framework to identify skin lesion(s) in dermoscopy images. Unlike the conventional density-based clustering algorithms, proposed algorithm expands regions only at borders of a cluster that in turn speeds up the process without losing precision or recall. In our method, border regions are represented with one or more simple polygons at any time. We tested our algorithm on a dataset of 100 dermoscopy cases with multiple physicians’ drawn ground truth borders. The results show that border error and f-measure of assessment averages out at 6.9% and 0.86 respectively. © 2010 Elsevier Ltd. All rights reserved. 1. Introduction Melanoma is the fifth most common malignancy in the United States [1]. Malignant melanoma, the most deadly form of skin can- cer, is one of the most rapidly increasing cancers in the world. 8441 deaths out of 68,720 incidences are estimated numbers in the United States during 2009 [2]. Early diagnosis is particularly important for melanoma since it can be cured with a simple excision operation in early stages of the disease. Dermoscopy, which is one of the non-invasive skin imag- ing techniques, has become a principal tool in the diagnosis of melanoma and other pigmented skin lesions. It involves optical magnification of the region-of-interest, which makes subsurface structures more visible than conventional macroscopic images [3]. This in turn improves screening characteristics and provides greater differentiation between difficult lesions such as pigmented Spitz nevi and small, clinically equivocal lesions [4]. However, it has also been demonstrated that dermoscopy may actually lower the diagnostic accuracy in the hands of inexperienced dermatologists [5]. Therefore, novel computerized image understanding tools are needed to minimize the diagnostic errors. These errors are gener- ally caused by the complexity of the incidents and the subjectivity of visual interpretations [6,7]. ∗ Corresponding author. Tel.: +1 9038865497. E-mail address: mutlu mete@tamu-commerce.edu (M. Mete). For many reasons, delineation of region-of-interest is the first and key step in the computerized analysis of skin lesion images. First of all, the border structure provides essential information for an accurate diagnosis. For instance, asymmetry, border irregularity, and abrupt border cutoff are some of the critical features calcu- lated based on the lesion border. Furthermore, the extraction of other critical clinical indicators such as atypical pigment networks, globules, and blue-white areas depend on the border detection [8]. In the literature, many algorithms were proposed to detect the borders in dermoscopy images. Those include the principal component transform (PCT)/median cut algorithm [9], adaptive thresholding, the first image plane of the PCT [10], thresholding in the blue image plane [11], k-means clustering [12], split-and-merge [9,13], a segmentation technique based on a Markov random field (MRF) image model [14], and a non-linear diffusion technique [12]. Schmid [15] proposed an algorithm based on color clustering. First, a two-dimensional histogram is calculated from the first two principal components of the CIE L*u*v* color space. The histogram is then smoothed and initial cluster centers are obtained from the peaks using a perceptron classifier. At the final step, the lesion image is segmented by using a modified version of the fuzzy c- means clustering algorithm. Gao et al. [12] created two methods: one based on stabilized inverse diffusion equations, a form of non- linear diffusion and another one based on Markov random fields in which the model parameters are estimated using the mean field theory. Regarding boundary of clusters, Lee and Estivill-Castro [16] introduced a new algorithm of polygonization based on bound- 0895-6111/$ – see front matter © 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.compmedimag.2010.07.007