融合通道注意力与残差密集的U-Net视网膜血管分割

U-Net retinal vessel segmentation based on fusion of channel attention and residuals

视网膜血管的精准分割是辅助眼科医生诊断和大规模眼科疾病自动筛查的重要前提,已成为临床的迫切需求。针对现有视网膜细小血管分割不足以及精确度有待提高等问题,提出了一种融合通道注意力机制与残差密集连接模块的改进型U-Net算法,先利用通道注意机制来增强网络的识别能力,再利用残差密集模块代替传统的卷积模块来提升网络分割细小血管的性能。在DRIVE和CHASE数据集上的实验结果表明,与其他算法相比,该算法的ACC、SE、SP和AUC值均比较高,分割效果较好。

accurate segmentation of retinal vessels is an important prerequisite for assisting ophthalmologists in diagnosis and large-scale automatic screening of ophthalmic diseases, and has become an urgent need in clinic. In view of the shortcomings of the existing retinal small blood vessel segmentation and the problem that the accuracy needs to be improved, an improved U-Net algorithm combining the channel attention mechanism and the residual dense connection module is proposed. First, the channel attention mechanism is used to enhance the recognition ability of the network, and then the residual dense module is used to replace the traditional convolution module to improve the performance of the network in segmenting small blood vessels. The experimental results on drive and chase data sets show that the algorithm has higher ACC, SE, SP and AUC values and better segmentation effect compared with other algorithms.