Regularized UNet for Automated Pancreas Segmentation
Jia Fan, Xue‐Cheng Tai · 2019
Leading other traditional methods by a large margin, convolutional neural networks (CNNs) become the first choice for dense classification problems such as semantic segmentation. However, when given few training images, CNNs often could not deal with details well. Coarse edges and isolated points often appear in the segmentation results provided by CNNs. Since medical datasets usually contain dozens to hundreds training samples, the segmentation results need further refinement. In this paper, we implement regularized UNet (RUNet) with multi-step primal-dual block which is an end-to-end framework to regularize the segmentation results. The proposed framework could produce smooth edges and eliminate isolated points. Comparing to other post-processing methods, our method needs little extra computation thus is effective and efficient.