Analysis of common melanomas and nevi using geometrical features

Abstract

This study aims to use some geometrical features to investigate and classify two common skin lesions, nevi and melanomas. Two sets of malignant and benign lesions were investigated; the first set contains 70 images from melanoma-diagnosed patients and the second set contains 100 of regular nevi images. A two-steps processing method is followed. First, segmentation of the skin lesions based on the optimal threshold determined by the histogram is performed, then a clustering operation using the Fuzzy C-Mean algorithm clusters the geometrical features. Asymmetry index, compactness, circularity and eccentricity were proposed as significant geometrical features. To validate the clustering results, the Euclidean distance between centroids was computed. Eccentricity well separates all objects under analysis and can differentiate between selected classes of skin lesions.