A Deep Convolutional Neural Network for Melanoma Recognition in Dermoscopy Images

Abstract

Automated melanoma recognition in dermoscopy images is a challenging task due to a set of hindrances including low contrast skin images, the resemblance of melanoma and non-melanoma skin lesions, and the great variety in this type of skin cancer. However, in this study, a fully automated method is proposed which recognizes the melanoma lesions from the non-melanoma lesions with high accuracy. Convolutional Neural Networks (CNNs) have made great strides in the field of recognition and classification of medical images. Based on this ground, a deep convolutional neural network is proposed that acts as the central pillar of the proposed melanoma recognition method. In order to compensate for the lack of training data, data augmentation techniques have been employed. The proposed method is a merger of the features elicited from the proposed Convolutional Neural Network architecture and a Support Vector Machine (SVM) classifier. The classifier categorizes the input dermoscopy images into two main classes of Melanoma and non-Melanoma skin lesion images with a promising accuracy of 89.52\%, which outperforms the state-of-art methods.