Image Pattern Analysis towards Classification of Skin Cancer through Dermoscopic Images

Malignant melanoma is one of the most dangerous forms of skin cancer. Death rates are quite high because of their proliferation ability to other body parts. Melanoma can be cured with early and correct diagnosis therefore there is a significant requirement of a system that can correctly differentiate benign and melanoma lesions. A computer-aided methodology for the classification of benign and melanoma is proposed in this study. To determine the area of interest of the lesion part, dermoscopic skin lesion images are first segmented using k-means clustering. Next, extensive feature extraction is performed on segmented images by using feature descriptors: local binary patterns (LBP), histogram of oriented gradients (HOG), and bag of visual words (BoVW). These features are evaluated on a wide range of classification methodologies. Experimental analysis reveals that BoVwfeatures with support vector machines classifier yield the highest results in terms of 99.8% accuracy, 100% sensitivity, and 99.5% specificity.