Two-stream convolutional networks for skin cancer classification

Skin cancer has become a popular disease as it represents one-third of all diagnosed cancers around the world. This cancer kills many people every year, and it is therefore important to detect skin cancer at an early stage to increase the survival rate. Detecting the exact type of skin cancer is a challenging problem due to the similar appearance between types of skin cancer. In this paper, we propose a novel approach based on combining two streams of convolutional neural networks to detect and classify skin cancer. We use RGB images and gradient images as inputs to the two streams of convolutional neural networks. We combine gradient images that contain high frequency information with RGB images to enhance classification accuracy. The experimental results show that the proposed approach achieves state-of-the-art accuracy (96.67%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$96.67\%$$\end{document}) using Human Against Machine (HAM10000) data set.