Analysis of retinal vessel networks using quantitative descriptors of vascular morphology

Abnormalities in the vascular pattern of a retina, such as morphologic changes in vessel shape, branching pattern, width, tortuosity, or the appearance of retinal lesions, may be associated with the occurrence of retinopathies or cardiovascular diseases. Thus, an automated quantitative analysis of changes in vessel morphology may help indicating the clinical signs of aforementioned retinopathies, describing their early occurrence or severity. The responses obtained from different types of retinal vessels, i.e., arteries and veins, may be variable to retinopathies and their measurement may lead to a more precise diagnosis compared to that by the average response accounted for the entire vessel network.

I propose a set of automated methods in order to analyze the retinal vessel network and to quantify its morphologic properties with respect to arteries and veins, in two-dimensional color fundus images. The analytical methods include; 1) Forma- tion of a well connected vessel network, 2) Structural mapping of a vessel network, 3) Artery-venous classification, and 4) Blood vessel hemorrhage detection. The quan- tification methods include vessel morphology analysis based on the measurement of tortuosity, width, branching angle, branching coefficient, and fractal dimension. The aforementioned morphologic parameters are measured with respect to arteries and veins separately in a vessel network. The methods are validated with the manually annotated retinal fundus images as a ground truth.

The major contribution of this thesis includes the development of automated methods for; 1) Identification and separation of retinal vessel trees for individual vessel analysis, 2) Automated quantification of morphologic characteristics of retinal vessels for quick and precise measurement, 3) Automated quantification of vessel morphology with respect to arteries and veins, and 4) Analysis of two datasets, a) malarial retinopathy subject dataset, b) longitudinal study dataset. The ability of the automated methods to quantify the retinal vessel specific properties may enable the individual vessel analysis as an alternative to a time- consuming and subjective clinical evaluation, or to a quantitative morphology char- acterization averaged over the entire vessel network. The objective evaluation may indicate the progression of retinopathies precisely and may help characterizing nor- mal and abnormal vascular patterns with respect to arteries and veins. This may enable a quick diagnosis, treatment availability, prognosis, and facilitation of clinical health-care procedures in remote areas.