摘要
眼底视网膜是唯一可用肉眼直接并集中观察到动脉、静脉与毛细血管的部位,因而眼底图像成为医生诊断眼底疾病及糖尿病、高血压、高血脂等疾病的重要依据.高质量的眼底图像是医生对眼底疾病患者进行病情诊断与治疗的前提.根据眼底相机采集到的视网膜图像中眼底结构清晰度、图像对比度等条件对眼底图像质量进行分类成为一个既具有研究价值又极具挑战性的难点问题.首先简述了眼底图像质量分类的研究意义和实用价值,回顾了其发展历史;然后介绍了方法分类、每类方法的基本思想并梳理了各类方法中代表性算法及其特点;之后针对用于眼底质量分类的数据集,分析比较了主要眼底图像质量分类方法的性能.分析表明,传统方法中依据眼底结构特征判断视网膜图像质量相较于通用图像质量参数更加客观,而随着神经网络与机器学习的出现,在大数据驱动下,基于卷积神经网络的质量分类方法在准确率与鲁棒性方面性能更佳.最后对眼底图像质量分类未来的发展趋势进行展望.
Fundus is the only part where the arteries, veins and capillaries can be observed with the naked eyes directly and centrally. Therefore, the fundus image is an important basis for doctor to diagnose fundus diseases and some other diseases such as diabetes, hypertension, and hyperlipidemia. High quality fundus images are the premise for doctors to analyze and treat the fundus diseases for patients. Classification of fundus image quality based on conditions of fundus structure clarity and image contrast in retinal images collected by fundus cameras has become a difficult problem with both research value and challenge. Firstly, the research significance and practical value of fundus image quality classification are briefly described, and its development history is reviewed. Secondly, the method classification and the basic idea of each method are introduced and the representative algorithms and their characteristics in various methods are introduced. Thirdly, the data set for fundus image quality classification is introduced, and the performance of the main fundus image quality classification methods is analyzed and compared. The analysis shows that among the traditional method, it is more objective to judge the quality of retinal image based on the characteristics of fundus structure than general image quality parameters. With the emergence of neural network and machine learning, the quality classification method based on convolutional neural network driven by big data has better performance in accuracy and robustness. Finally, the future development trend of fundus image quality classification is prospected.
作者
张芳
赵东旭
肖志涛
徐旭
耿磊
吴骏
刘彦北
王雯
Zhang Fang;Zhao Dongxu;Xiao Zhitao;Xu Xu;Geng Lei;Wu Jun;Liu Yanbei;Wang Wen(School of Electronics and Information Engineering,Tianjin Polytechnic University,Tianjin 300387;Tianjin Key Laboratory of Optoelectronic Detection Technology and System,Tianjin 300387)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2020年第3期501-512,共12页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61601325)
天津市科技重大专项与工程项目(17ZXSCSY00060,17ZXHLSY00040)
天津市高等学校创新团队培养计划项目(TD13-5034).
关键词
眼底图像质量分类
通用图像质量参数
眼底结构信息
深度学习
fundus image quality classification
generic image quality parameters
fundus images structural information
deep learning
