视网膜血管提取算法  

Algorithms for extracting retinal blood vessels

作　　者：李雪 马瑜 郭姝琪 王鹏志 LI Xue;MA Yu;GUO Shu-qi;WANG Peng-zhi(School of Physics and Electrical Engineering,Ningxia University,Yinchuan 750021,China)

机构地区：[1]宁夏大学物理与电子电气工程学院,宁夏银川750021

出　　处：《计算机工程与设计》2024年第12期3786-3793,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62041108)。

摘　　要：为提高眼底视网膜血管分割的精确度,提出一种RAIterNet级联视网膜血管分割追踪算法,在预处理与分割网络模型阶段分别进行图像增强与精度提升。使用自适应分数阶次微分对待分割数据集进行增强提高血管图像质量,提升待分割图像血管与背景之间的对比度,利用RAIterNet模型对视网膜进行分割,基于连续性追踪方法对微弱结构的末端毛细血管进行追踪,可有效分割视网膜血管中难以分割的微细血管。算法在数据集DRIVE上进行测试,实验结果表明,Acc能够达到0.9650,F1分数为0.9006,获得了0.9807的AU-ROC曲线下面积,主观和客观结果验证了算法的有效性。To improve the accuracy of retinal blood vessel segmentation,an improved RAIterNet cascaded retinal blood vessel segmentation and tracking algorithm was proposed,and the image enhancement and accuracy improvement were performed respectively in the preprocessing and segmentation network model stages.The adaptive fractional differential was used to enhance the data set to be segmented to improve the quality of the blood vessel image and the contrast between the blood vessel and the background of the image to be segmented.The improved RAIterNet model was used to segment the retina.The end capillaries of the weak structure were tracked based on the continuous tracking method,which effectively segmented the tiny blood vessels in the retina that were difficult to segment.The algorithm was tested on the data set DRIVE.Experimental results show that the Acc can reach 0.9650,the F1 score is 0.9006,and the area under the AU-ROC curve of 0.9807 is obtained.The subjective and objective results verify the effectiveness of the algorithm.

关 键 词：视网膜血管提取 卷积神经网络 递归残差卷积 注意力机制 分数阶微分 自适应分数阶增强 血管追踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]