Deep Convolutional Pixel-wise Labeling for Skin Lesion Image Segmentation

Melanoma is one of the deadliest form of cancer with an increasing incidence rate. The development of automatic diagnostic tools for the early detection of skin cancer lesions in dermoscopic images can help to reduce melanoma-induced mortality. In this paper, we present an automatic method for skin lesion image segmentation based on a deep learning algorithm for pixel-wise labeling. Experimental results have been obtained by testing two network architectures on publicly available data and, in order to show that the used approach is not data set related, we have used the ISIC database for training the network and the PH2 database for testing. The results show that the proposed approach achieves a very accurate segmentation even in presence of hair and air/oil bubbles. An additional contribution of this work is the development of a semi-automatic GUI for data annotation that can be used to generate more test images.