RefineU-Net: Improved U-Net with Progressive Global Feedbacks and Residual Attention Guided Local Refinement for Medical Image Segmentation

•A novel FCN architecture called RefineU-Net is proposed to improve the performance of U-Net on medical image segmentation.•A global refinement module is proposed to generate intermediate layers in skip connections to alleviate semantic gap problems.•A local refinement module is proposed using a residual attention gate to generate discriminative attentive features.•The proposed RefineU-Net outperforms multiple U-Net based methods on four public datasets of medical segmentation.