Skin Lesion Segmentation by U-Net with Adaptive Skip Connection and Structural Awareness

Skin lesion segmentation is one of the pivotal stages in diagnosis melanoma. Many methods have been proposed but, to date, this still a challenging task. Variations size and color, fuzzy boundary low contrast between normal skin are adverse factors for deficient or excessive delineation lesions, even inaccurate location detection. In paper, counter these problems, we introduce deep learning method based on U-Net architecture, which performs three tasks, namely segmentation, distance map regression contour The two auxiliary tasks provide an awareness shape main encoder, improves object localization pixel-wise classification transition region from tissues healthy tissues. Moreover, concerning large variation size, Selective Kernel modules, placed skip connections, transfer multi-receptive field features encoder decoder. Our evaluated publicly available datasets: ISBI2016, ISBI 2017 PH2. extensive experimental results show effectiveness task segmentation.