Retinal microvasculature is known to be affected by systemic and cardiovascular diseases along with common eye diseases.
1–3 Quantitative fundus photography is essential for screening, diagnosis, and treatment assessment of eye diseases. Objective and automated classification of vasculature distortions in fundus images holds great potential to help physician decision making, foster telemedicine, and explore early screening of eye diseases in primary care environments.
4
Different diseases and progressing stages may affect arteries and veins differently. For example, arterial narrowing is a well-established phenomenon associated with hypertension, whereas venous widening is associated with stroke and cardiovascular diseases.
5–8 Artery–vein (A-V) caliber ratio has been used as a predictor of these diseases.
3,5,6,9,10 However, manual differentiation of arteries or veins is time consuming. Therefore, a number of algorithms have been proposed to explore computer-aided classification of A-V vessels.
11–20 Computer-aided classification typically requires multiple steps: extracting the blood vessel map, differentiating the artery and vein, quantifying any changes, and assessing the condition of the patient. Most of the vessel classification algorithms are based on the color and intensity information of arteries and veins.
13–16,18–20 Because of the presence of oxygenated blood, the arteries have lighter color intensity. However, this difference becomes less significant as the blood vessels propagate toward the fovea due to intensity and contrast variability. Therefore, the small vessels have to be tracked to their origin near the optic disk to be classified reliably.
21 Semiautomatic algorithms
11,12,17 for supervised classification have also been proposed based on vessel-tracking techniques. In supervised classification,
16,22,23 the intra- and interimage light variation make it quite challenging to get high accuracy in A-V classification. Furthermore, these algorithms require a large number of training sets with manual annotations from clinicians. Some researchers have tried incorporating functional features such as optical density ratio (ODR) to identify arteries and veins in dual-wavelength images obtained in red and green channels.
23–25 However, high sensitivity was achieved only for large vessels, leaving reliable A-V classification difficult for small vessels at the macular area that are vulnerable to many eye diseases.
In this work, we introduce an automated method that combines ODR analysis and a blood vessel tracking (BVT) algorithm to enable A-V classification at arteriole and venule level. As a functional feature, ODR is used to identify arteries and veins near the optic disk, while a BVT algorithm maps veins or arteries from source to endpoint using vessel curvature and angle information. Incorporating a vessel-enhancement algorithm with the tracking algorithm allows reliable A-V classification. We implemented the method on 50 color fundus images from 35 nonproliferative diabetic retinopathy (NPDR) patients and validated the results by comparing them to manual annotation from two independent observers. The classification performance is validated using sensitivity, specificity, and accuracy metrics along with graphical metrics, that is, a receiver operation characteristics (ROC) curve. We also measured two quantitative features, A-V caliber and tortuosity, to evaluate the effect of hypertension on the retinal vessels of NPDR patients.