Modified watershed technique and post-processing for segmentation of skin lesions in dermoscopy images

✍ Scribed by Hanzheng Wang; Randy H. Moss; Xiaohe Chen; R. Joe Stanley; William V. Stoecker; M. Emre Celebi; Joseph M. Malters; James M. Grichnik; Ashfaq A. Marghoob; Harold S. Rabinovitz; Scott W. Menzies; Thomas M. Szalapski

• Publisher: Elsevier Science
• Year: 2011
• Tongue: English
• Weight: 385 KB
• Volume: 35
• Category: Article
• ISSN: 0895-6111
• DOI: 10.1016/j.compmedimag.2010.09.006

✦ Synopsis

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.