【摘要】 U-Net在许多医学图像分割问题中具有先进的性能，因此提出了一种基于边缘增强和特征金字塔的U型分割网络并应用于视网膜血管分割。通过基于梯度算子的边缘增强模块获取额外的边缘先验，以无监督的方式增强边界特征和提高网络对细小血管的辨识能力，利用特征金字塔级联模块帮助网络提取更丰富的语义特征，并将传统卷积替换为Octave卷积方式以更好地提取特征。在公开的2个眼底图像数据集上进行实验，结果证明改进的方法具有更好的性能，有效地提高了分割结果中血管的完整性和连续性。更多还原

【Abstract】 U-Net provides the most advanced performance in many medical image segmentation problems. The U-shaped segmentation network is proposed, based on boundary enhancement and feature pyramid and applied to retinal vessel segmentation successfully. The proposed model obtains additional edge priors through the boundary enhancement module based on the gradient operator to enhance the boundary features and improve the network’s ability to identify small blood vessels in an unsupervised manner, and use the feature pyramid module to help the network extract richer semantic features. The traditional convolution is also replaced in the model with a new convolution method to extract better features. The experiment is carried out on two public fundus image data sets, the results prove that the proposed method has effectively improved the integrity and continuity of blood vessels in the segmentation.更多还原