Developing an automatic method for separation of arteries from veins in retinal images

Analyzing the retinal blood vessels can provide very helpful information to doctors for early detection of diseases such as diabetic retinopathy due to large number of patients. These diseases affect Blood vessels differently leading to thinner veins, thicker arteries or vice versa. Hence an abnormal width ratio of artery to vein (AVR) may be a sign of infection. To examine each class of blood vessels or measuring AVR, arteries and veins should be separated carefully. There have been a few researches done for classification of retinal Blood vessels up to this time. Some of their results are not comparable due to small and non standard databases used for evaluation of the respective methods. In this paper we focus to study appearance of vessels in different color spaces such as RGB and HSL to extract best features from inner and outer parts of vessels for classification of retinal blood vessels particularly the major ones. Since most of consecutive points in a vessel segments have similar features and these points belong to same type of vessel we attempt to divide long vessel segments into smaller ones and extract features for each vessel segment instead of every centerline points. Evaluating our method on DRIVE database we achieve 86% recognition rate on major vessels and 87.58% on pair main vessels in upper and lower region of retinal images, using a few sample points and a small feature set which decrease calculations noticeably. This work can help for future automatic tracking based methods to classify the smaller vessels or for determining the A/V ratio in major vessels.