基于SKA-Unet的眼底图像视网膜血管自动分割  

Automatic segmentation of retinal vessels in fundus images based on SKA-UNET

作　　者：王雪花 林静怡 吕德佳 许祥丛 韩定安[1,2] WANG Xue-hua;LIN Jing-yi;LV De-jia;XU Xiang-cong;HAN Ding-an(School of Physics and Optoelectronic Engineering,Foshan University,Foshan 528225,China;Guangdong-Hong Kong-Macao Joint Laboratory for Intelligent Micro-Nano Optoelectronic Technology,Foshan University,Foshan 528225,China)

机构地区：[1]佛山科学技术学院物理与光电工程学院,广东佛山528200 [2]粤港澳智能微纳光电技术联合实验室,广东佛山528225

出　　处：《佛山科学技术学院学报（自然科学版）》2023年第1期17-22,共6页Journal of Foshan University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61805038,62075042,61771139);粤港澳智能微纳光电技术联合实验室研究基金资助项目(2020B1212030010)。

摘　　要：针对视网膜血管结构复杂,对比度较低及边界模糊等问题,构建了一种改进的Unet模型来自动分割视网膜血管。该算法在Unet基础上引入了选择性内核和注意力机制,通过增加多重感受野得到目标多维度的特征信息,并将注意力集中在最有效的特征中,从而提升网络分割血管的性能。利用网上公开的DRIVE和STARE两个常用眼底图像数据集训练和评估该模型,得到骰子系数、精确度、特异性、灵敏度和准确率分别为86.49%、88.65%、98.82%、84.5%和97.46%。实验结果表明,该算法能有效地分割视网膜血管,获得血管的形态和结构信息,对于诊断和治疗眼底相关疾病具有重要意义。Aimed at the problems of complex retinal vascular structure,low contrast and blurred boundary in fundus images,an improved Unet model was constructed to segment retinal vessels automatically.This algorithm introduces a selective kernel and attention mechanism into a general Unet.It obtains multi-dimensional feature information of the target by adding multiple receptive fields,and focuses attention on the most effective feature,thus improving the performance of blood vessels’segmentation with the network.The model was trained and evaluated using two common fundus image datasets,DRIVE and STARE,and the dice coefficient,accuracy,specificity,sensitivity and accuracy of 86.49%,88.65%,98.82%,84.5%and 97.46%were obtained,respectively.The experimental results show that this algorithm can effectively segment the retinal blood vessels and obtain vascular morphology and structure information,which is of great significance for diagnosis and treatment of fundus related diseases.

关 键 词：Unet 视网膜血管 图像分割 选择性内核机制 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R774[自动化与计算机技术—计算机科学与技术]