Modified watershed technique and post-processing for segmentation of skin lesions in dermoscopy images
In previous research, a watershed-based algorithm was shown to be useful for automatic lesion segmentation in dermoscopy images, and was tested on a set of 100 benign and malignant melanoma images with the average of three sets of dermatologist-drawn borders used as the ground truth, resulting in an overall error of 15.98%. In this study, to reduce the border detection errors, a neural network classifier was utilized to improve the first-pass watershed segmentation; a novel "edge object value (EOV) threshold" method was used to remove large light blobs near the lesion boundary; and a noise removal procedure was applied to reduce the peninsula-shaped false-positive areas. As a result, an overall error of 11.09% was achieved.
H. Wang and X. Chen and R. J. Stanley and W. V. Stoecker and M. E. Celebi and J. M. Malters and J. M. Grichnik and A. A. Marghoob and H. S. Rabinovitz and S. W. Menzies and T. M. Szalapski and R. H. Moss, "Modified watershed technique and post-processing for segmentation of skin lesions in dermoscopy images," Computerized Medical Imaging and Graphics., Elsevier, Mar 2011.
The definitive version is available at http://dx.doi.org/10.1016/j.compmedimag.2010.09.006
Electrical and Computer Engineering
National Institute of Health (U.S.)
Keywords and Phrases
Malignant Melanoma; Neural Network; Segmentation; Watershed
Library of Congress Subject Headings
Article - Journal
© 2011 Elsevier, All rights reserved.