Automatic skin lesion analysis towards melanoma detection

Deep learning methods for image analysis have shown impressive performance in recent years. In this paper, we present deep learning based approaches to solve two problems in skin lesion analysis using a dermoscopic image containing skin tumor. In the first problem, we use a fully convolutional-deconvolutional architecture to automatically segment skin tumor from the surrounding skin. In the second problem, we use a simple convolutional neural network and VGG-16 architecture using transfer learning to address the two different tasks in skin tumor classification. The proposed models are trained and evaluated on standard benchmark datasets from the International Skin Imaging Collaboration (ISIC) 2017 Challenge, which consists of 2000 training samples and 600 testing samples. The result shows that the proposed methods achieve promising performances. In the first problem, the average value of Jaccard index for lesion segmentation using fully convolutional-deconvolutional architecture is 0.507. In the second problem, the values of area under the receiver operating characteristic curve (AUC) on two different lesion classifications using VGG16 with transfer learning are 0.763 and 0.869, respectively; the average value of AUC in two tasks is 0.816.