Skin Melanoma Classification Using Deep Convolutional Neural Networks

Abstract

This chapter assesses the existing methods for skin lesion classification, especially deep learning-based methods. It proposes two models using the theory of transfer learning for the Alex-net and Resnet50 architecture to classify tumors as melanoma or not-melanoma. The classification layer has been replaced by the multiclass support vector machine, where a fast-linear solver is compatible with the two classes of skin images: melanoma and nevus. S. A. Kostopoulos et al. proposed a computer-based analysis of plain photography using a probabilistic neural network to extract the features and decide if the lesion is melanocytic or melanoma. K. Bunte et al. proposed a classifying melanoma system. In addition to first, second, and third moment orders, Bunte and his coauthors used correlograms, coherence vectors, and histograms for the representation of color image features. Various kinds of skin tumor have been found, like squamous cell carcinoma, basal cell carcinoma, and melanoma; the last of these is the most unpredictable.