Feature extraction of dermatoscopic images by iterative segmentation algorithm

摘要

Abstract. Since the introduction of epiluminescence microscopy (ELM), image analysis tools have been extended to the fieldnof dermatology, as an attempt to algorithmically reproduce clinical evaluation. Accurate image segmentation of skin lesionsnis one of the key steps for useful, early, and non-invasive diagnosis of coetaneous melanomas. This paper proposes an imagensegmentation algorithm to extract the true border that reveals the global structure irregularity (indentations and protrusions),nwhich may suggest excessive cell growth or regression of a melanoma. The algorithm is applied to the blue channel of the RGBncolour vectors to distinguish lesions from the skin and. Analysis of image background is applied by recursive measure of thenmedian and standard deviation of background. This will facilitate automatic and recurring noise reduction and enhancement bynimage pre-processing. The algorithm also does not depend on the use of rigid threshold values, because an optimal thresholdingnalgorithm “isodata algorithm" that is used determines an optimal threshold iteratively. Experiments are performed on diversitynof synthetic skin images that model real hair and lesions of different border irregularities. The aim is to verify the capability ofnthe segmentation algorithm in extracting and characterizing the true features of the processed skin lesions. The next phase ofntest applies the algorithm to real skin lesions representing high resolution ELM images. We demonstrate that we can enhancenand delineate pigmented networks in skin lesions visually, and make them accessible for further analysis and classification.