Improving Dermoscopic Image Segmentation with Enhanced Convolutional-Deconvolutional Networks

摘要

Automatic skin lesion segmentation on dermoscopic images is an essential stepin computer-aided diagnosis of melanoma. However, this task is challenging dueto significant variations of lesion appearances across different patients. Thischallenge is further exacerbated when dealing with a large amount of imagedata. In this paper, we extended our previous work by developing a deepernetwork architecture with smaller kernels to enhance its discriminant capacity.In addition, we explicitly included color information from multiple colorspaces to facilitate network training and thus to further improve thesegmentation performance. We extensively evaluated our method on the ISBI 2017skin lesion segmentation challenge. By training with the 2000 challengetraining images, our method achieved an average Jaccard Index (JA) of 0.765 onthe 600 challenge testing images, which ranked itself in the first place in thechallenge