Vessel Extraction in Fluorescein Angiograms of the Human Retina Using a Supervised Classifier

Fluorescein angiograms of the human retina are widely used in the diagnosis and treatment of several diseases such as diabetic retinopathy and age relate macular degeneration. They analyze the micro circulation of the retina and choroid. Hence, accurate extraction of the vascular tree network is of crucial importance. Previous approaches to retinal vessel extraction assume either bright vessels on a dark background or dark vessels on a lighter background. We present a supervised method to segment retinal blood vessels from fluorescein angiograms images considering that both dark and bright vessels can be seen on an even background within the same image. The proposed approach is based on the eigenvalue decomposition of the Hessian matrix and Fisher's linear discriminant analysis. This technique was implemented and tested using retinal fundus images and images from a fluoresceing angiogram in order to evaluate its performance, moreover, squared error metric, performance and true positive ratio were computed. The approach, in most of the cases, was able to differentiate between dark vessels and bright vessels' edges.