A Brief Analysis of U-Net and Mask R-CNN for Skin Lesion Segmentation

Abstract

A brief analysis on the use of two deep neural architectures, the U-Net and Mask R-CNN for the segmentation of skin lesions in dermoscopic images is presented. The two systems were adapted to use the dataset provided by the International Skin Imaging Collaboration (ISIC) for its 2017 challenge and different experiments were carried out. Results showed that the Mask-R-CNN obtained better performance than U-Net, also with lower computation times, being a feasible architecture to further analysis and application also to skin lesion classification.