摘要
视网膜血管自动分割是实现糖尿病、心血管疾病,以及多种眼科疾病自助诊断的关键步骤。血管末端消失、血管与视盘相混淆是导致分割准确率下降的主要原因。针对此类问题,提出一种基于"编码器-解码器"的复合特征视网膜血管分割方法。将U-Net的对称结构与HRNet保持高分辨率表征的方法相结合,构建复合特征提取融合模块,使神经网络充分捕捉图像中的浅层信息,提高视网膜分割准确率。在公开DRIVE数据集上进行实验,模型在测试集上的准确率、特异性、灵敏度、AUROC和AUCPR分别达到0.953 7、0.984 6、0.764 4、0.983 0、0.923 9。实验结果表明,相对于目前主流的视网膜血管分割方法,所提方法表现出了一定的优势。
Automated retinal vessel segmentation is a key step in the self-diagnosis of diabetes,cardiovascular disease and various ophthalmic diseases.The disappearance of blood vessels and the confusion of blood vessels and optic discs are the main reasons for the decline in segmentation accuracy.Aiming at these problems,a retinal vessel segmentation method of composite feature based on“encoder-decoder”is proposed.It combined the symmetric structure of U-Net with the method of high-resolution characterization of HRNet to construct a composite feature extraction fusion module.The neural network could fully capture shallow information in the image,and the accuracy of retinal segmentation was improved.Experiments were performed on the public DRIVE dataset.The accuracy,specificity,sensitivity,AUROC and AUCPR scores of the model on the test set reached 0.9537,0.9846,0.7644,0.9830,and 0.9239.The experimental results show that compared with the current mainstream retinal vessel segmentation methods,the proposed method shows its advantages.
作者
孟颖
田启川
吴施瑶
Meng Ying;Tian Qichuan;Wu Shiyao(School of Electrical and Information Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Key Laboratory of Intelligent Processing for Building Big Data,Beijing 100044,China)
出处
《计算机应用与软件》
北大核心
2021年第8期227-232,267,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61871021)
北京建筑大学研究生创新项目(PG2019074)。
