Texture Classification of skin lesion using convolutional neural network

Melanoma skin cancer is becoming more and more prevalent. Fortunately, early detection can significantly increase the chances of a cured patient. However, an accurate recognition of this type of cancer is challenging due to the difficulty to distinguish the difference between melanoma and non-melanoma skin cancer. Recently, many approaches have been proposed to classify skin lesion based on deep learning. In this paper, an improved skin lesion detection and classification using convolutional neural network is proposed. The main idea is to use only lesion textural information as CNN input instead of the whole images. The textural lesion is obtained by the projection of the segmented object and texture components obtained by the use of the multi-scale decomposition. Experimental results show that the proposed approach got a higher accuracy compared to other approaches form literature.