Improvement on recognition of major arteries and veins on retinal fundus images by template matching with vessel models

DOI
  • MURAMATSU Chisako
    Department of Intelligent Image Information, Graduate School of Medicine, Gifu University
  • MIZUKAMI Atsuki
    Department of Intelligent Image Information, Graduate School of Medicine, Gifu University
  • HATANAKA Yuji
    Department of Electronic Systems, School of Engineering, the University of Shiga Prefecture
  • SAWADA Akira
    Department of Intelligent Image Information, Graduate School of Medicine, Gifu University
  • HARA Takeshi
    Department of Intelligent Image Information, Graduate School of Medicine, Gifu University
  • YAMAMOTO Tetsuya
    Department of Ophthalmology, Graduate School of Medicine, Gifu University
  • FUJITA Hiroshi
    Department of Intelligent Image Information, Graduate School of Medicine, Gifu University

Bibliographic Information

Other Title
  • 血管モデルのテンプレートマッチングによる眼底画像上の主幹動静脈認識精度の改善

Search this article

Abstract

Studies suggest association of retinal microvascular abnormalities with cardiovascular and cerebrovascular diseases. Arteriolar narrowing, which can be assessed by arteriolar-to-venular diameter ratio (AVR) on retinal fundus images, is one of the findings for hypertensive retinopathy. We have been studying an automated method for measuring AVR in hope of improving diagnostic efficiency and consistency of ophthalmologists. One of the problems in our previous method was that the suboptimal segmentation accuracy of the major arteries, especially those with low contrast and central reflex. In order to improve the recognition rate of major vessel pairs, synthetic vessel models were created, and the missed or broken arteries were identified by template matching. The method was applied to 22 retinal fundus images, including cases with arteriolar narrowing. By use of the models with 2 different shape profiles and various sizes, the major vessel recognition rate was improved from 72.7% to 90.9%. The proposed method may be useful in automated measurement of AVR.

Journal

Related Projects

See more

Keywords

Details 詳細情報について

Report a problem

Back to top