王皓浩1,
孙刚1,
张谱2,
刘敏1,
高玲2
1.湖南大学电气与信息工程学院 ??长沙 ??410082
2.中南大学湘雅附属第二医院 ??长沙 ??410011
详细信息
作者简介:刘小燕：女，1973年生，教授，博士生导师，研究方向为图像处理技术及其应用、智能建模与控制
王皓浩：男，1994年生，硕士生，研究方向为医学图像处理技术
孙刚：男，1992年生，博士生，研究方向为医学图像处理技术
张谱：男，1985年生，博士生，研究方向为视网膜、脉络膜及玻璃体相关疾病
刘敏：男，1981年生，副教授，博士生导师，研究方向为计算机视觉、模式识别以及机器学习
高玲：女，1968年生，主任医师，研究方向为视网膜、脉络膜及玻璃体相关疾病
通讯作者:刘小燕 ? xiaoyan.liu@hnu.edu.cn
中图分类号:TP391; R445计量
文章访问数:1271
HTML全文浏览量:488
PDF下载量:33
被引次数:0
出版历程
收稿日期:2017-09-14
修回日期:2018-05-09
网络出版日期:2018-06-07
刊出日期:2018-08-01
A Novel Automatic Registration Method for Fluorescein Fundus Angiography Sequences Based on Mutual Information
Xiaoyan LIU1,,,Haohao WANG1,
Gang SUN1,
Pu ZHANG2,
Min LIU1,
Ling GAO2
1. College of Electrical and Information Engineering, Hunan University, Changsha 410082, China
2. Department of Ophthalmology, the Second Xiangya hospital of Central South University, Changsha 410011, China
摘要
摘要:荧光素眼底血管造影技术(FFA)是眼底疾病诊断的金标准，但是造影过程中病人不可避免地转动眼球，造成FFA图像序列中感兴趣区域(例如视网膜血管分支、新生血管)的位置发生变化，给后续的图像定量分析与病情准确评估诊断带来困难。针对上述问题，该文提出一种基于互信息的FFA图像序列配准方法。首先采用多尺度线性滤波方法分割出图像中的血管，并利用图像金字塔对分割后的图像进行下采样，然后利用互信息计算待配准图像与参考图像的相似性，通过进化策略对配准参数进行优化，获得互信息最大时图像的空间变换矩阵，实现FFA图像的配准。采用上述方法，对4位患者共计1039帧FFA图像进行测试，总体配准率达到93%，失败率仅为1%；与常用的配准方法相比，所提方法的配准率、配准速度和鲁棒性等综合性能良好，为FFA影像的定量分析在未来的临床应用奠定了基础。
关键词:荧光素眼底血管造影/
图像序列/
互信息/
血管分割
Abstract:Fluorescein Fundus Angiography (FFA) is regarded as the golden diagnostic criteria for fundus diseases. However, dislocation or rotation of the interested images on anatomic landmark (like retinal vascular branches, neovascularization), caused by inevitable eyeball movement, brings about difficulties in subsequent quantitative analysis and progress assessment of the diseases. In order to solve the above problems, a novel method based on mutual information is proposed for automatic registration of FFA image sequence. Firstly, the vessels of image sequence are segmented by multi-scale linear filter and down sampled hereafter by image pyramid. Then, the similarity of sampled images is calculated by mutual information and the evolution strategy is adopted to optimize the registration parameters. Finally, the transformation matrix with maximum mutual information is obtained to register the FFA image. Tests with FFA image sequences of 4 patients (total 1039 frames) show that the overall registration rate of the algorithm reaches 93% and the failure rate is only 1%. Compared with the classical registration methods, the proposed method shows better comprehensive performance in terms of registration rate, computing speed as well as robustness. It lays basic foundations for quantitative analysis on FFA images and potential clinical application.
Key words:Fluorescein Fundus Angiography (FFA)/
Image sequence/
Mutual information/
Vascular segmentation
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=f810a7f6-b62e-4f6e-9235-4dbfa2e6025a
