Detection of Melanoma Skin Lesions Using Phylogeny

Abstract

Asymmetry, boundary irregularity, color variegation, and diameter are popular clinical guidelines used for early detection of melanoma. Several researchers have proposed and framed methodologies to detect the melanoma using combined and separate features with the help of statistical theory, fuzzy logic, artificial neural networks, and machine learning techniques. When we combine all the techniques to quantify the four attributes (recently derived) within a single interface to diagnose melanoma skin lesion, it will be much complex and time consuming task. Hence, this paper proposes a novel methodology to detect melanoma skin cancer through phylogeny of pigmented skin lesions and it is a pilot scale attempt using the images of skin lesions. The construction of phylogeny has the following steps namely, the construction of image matrix, construction of image blocks, phylogenetic tree, and classification. Classification is implemented through average mean rank method as well as artificial neural networks. Performance of the proposed methodology is tested with 110 different images of skin lesions (BC cancer agency) and it is shown to predict 87 % successfully (melanoma/non-melanoma). The advantage of the proposed methodology is to predict the classification of melanoma skin lesions with the help of phylogeny achieves accuracy of 87 % with less time complexity. Some additional analysis is required for the remaining 13 % of images. Finally, it is proved that both average rank method and artificial neural network classification method predict almost the same accuracy.