Title: A primary morphological classifier for skin lesion images
Authors: Macatangay, Jules Matthew A.
Ruiz Jr., Conrado R.
Usatine, Richard P.
Citation: WSCG 2017: full papers proceedings: 25th International Conference in Central Europe on Computer Graphics, Visualization and Computer Visionin co-operation with EUROGRAPHICS Association, p. 55-64.
Issue Date: 2017
Publisher: Václav Skala - UNION Agency
Document type: konferenční příspěvek
URI: wscg.zcu.cz/WSCG2017/!!_CSRN-2701.pdf
ISBN: 978-80-86943-44-2
ISSN: 2464–4617 (print)
2464–4625 (CD-ROM)
Keywords: kožní léze;klasifikace;strojové učení;počítačové vidění
Keywords in different language: skin lesion;classification;machine learning;computer vision
Abstract: Classifying skin lesions, abnormal changes in skin, into their morphologies is the first step in diagnosing skin diseases. In dermatology, morphology is a categorization of a skin lesion’s structure and appearance. Rather than directly classifying skin diseases, this research aims to explore classifying skin lesion images into primary morphologies. For preprocessing, k-means clustering for image segmentation and illumination equalization were applied. Additionally, features utilized considered color, texture, and shape. For classification, k-Nearest Neighbors, Decision Trees, Multilayer Perceptron, and Support Vector Machines were used. To evaluate the prototype, 10-fold cross validation was applied over a dataset assembled from online resources. In experimentation, the morphologies considered were macule, nodule, papule, and plaque. Moreover, different feature subsets were tested through feature selection experiments. Experimental results on the 4-class and 3-class tests show that of the classifiers selected, Decision Trees were best, having a Cohen’s kappa of 0.503 and 0.558 respectively.
Rights: © Václav Skala - UNION Agency
Appears in Collections:WSCG 2017: Full Papers Proceedings

Files in This Item:
File Description SizeFormat 
Macatangay.pdfPlný text7,1 MBAdobe PDFView/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/29545

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.