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

ALBORNOZ, ENRIQUE M.; BOLAÑOS X.; CALDERÓN S.; ALFARO E.; MARTÍNEZ, CÉSAR E.

Budapest

Congreso; IEEE International Work Conference on Bioinspired Intelligence; 2019

IEEE Hungary Section

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