Residual-Attention UNet++: A Nested Residual-Attention U-Net for Medical Image Segmentation

Image segmentation is a basic technology in the field of image processing and computer vision. Medical an important application plays increasingly role clinical diagnosis treatment. Deep learning has made great progress medical segmentation. In this paper, we proposed Residual-Attention UNet++, which extension UNet++ model with residual unit attention mechanism. Firstly, improves degradation problem. Secondly, mechanism can increase weight target area suppress background irrelevant to task. Three datasets such as skin cancer, cell nuclei, coronary artery angiography were used validate model. The results showed that achieved superior evaluation scores Intersection over Union (IoU) 82.32%, dice coefficient 88.59% cancer dataset, 85.91%, IoU 87.74% nuclei dataset 72.48%, 66.57% dataset.