Automated independent extraction of major arteries and veins on retinal images

Arteriolar-to-venular diameter ratio(AVR)is measured for decision of retinal arteriolar narrowing, which is one of the findings for hypertensive retinopathy. We have been developing an automated method for measuring AVR to help diagnosis of ophthalmologists. For measuring AVR, arteries and veins must be segmented with high accuracy. Previous methods first extracted blood vessels and then classified arteries and veins by using linear discriminant analysis(LDA)of pixel-based-features. However, previous methods could not segment in cases with close contact between the artery and a vein. Therefore, this paper describes a novel method based on independent extraction of major arteries and veins. Veins were first extracted in the red channel of color retinal image. By removing the veins from the blood vessels extracted in the green channel of color retinal image, the arteries were obtained. Major veins were decided by using LDA proposed in the previous method. Major arteries were decided by using decision tree with three features. The method was applied to 22 retinal images including cases with arteriolar narrowing. As a result, 98% of major veins and 77% of major arteries were correctly identified. The proposed method may be useful in automated measurement of AVR.