Abstract:Background and objectives: Fluorescein angiography (FA) is widely used in ophthalmology for examining retinal hemodynamics and vascular morphology. Artery-venous classification is an important step in FA image processing for measurement of feature parameters, such as arterio-venous passage time (AVP) and arterio-venous width ratio (AVR) that are proven useful in clinical assessment of circulation disturbance and vessel abnormalities. However, manual artery-venous classification needs expertise and is rather time consuming, and little effort has been devoted to develop automatic classification methods. In order to solve this problem, we propose a novel artery-venous classification method using region growing strategy with sequential and structural features (RGSS). Methods: The main procedures of our proposed RGSS method include: (i) registration of FA image sequence by mutual-information method; (ii) extraction of sequential features of the dye perfusion process from the registrated FA images; (iii) extraction of vessel structural features from vascular centerline map; (iv) based on the obtained features, seeds of arteries and veins within initial growing region (here optic disk) are generated and then propagated in the entire vessel network using region growing strategy. The RGSS method was tested on our own dataset and public Duke dataset, and its performance was evaluated quantitatively. Results: Tests show that RGSS method is able to classify arteries and veins from the complicated vessel network in FA images, with high classification accuracy of 0.91 ± 0.04 on Duke dataset and 0.92 ± 0.03 on our dataset. The employed sequential and structural features are demonstrated to be effective in classifying thin arteries and veins at vessel crossings. Conclusions: Automatic artery-venous classification can be accomplished using our proposed RGSS method with high accuracy. The RGSS method not only emancipates ophthalmologists from hard work of manual marking of arteries and veins, but also helps in measuring important parameters (such as AVP and AVR) for clinical assessment of circulation disturbance and vessel abnormalities.

An Improved Arteriovenous Classification Method for the Early Diagnostics of Various Diseases in Retinal Image

Automated Artery-Vein Classification in Fundus Color Images.

Artery/vein automatic classification in retinal images and vessel diameter Measurement

The Arteriovenous Classification in Retinal Images by U-net and Tracking Algorithm

Automatic artery/vein classification methods for retinal blood vessel: A Review

Automatic retinal image diagnosis system for mass health screenings

An Artery/Vein Classification Method Based On Color And Vascular Structure Information

Multi-scale Interactive Network with Artery/Vein Discriminator for Retinal Vessel Classification

Supervised Vessels Classification Based on Feature Selection

Joint segmentation and classification of retinal arteries/veins from fundus images

An Improved U-Net Architecture for Simultaneous Arteriole and Venule Segmentation in Fundus Image.

Optimization of Retinal Artery/vein Classification Based on Vascular Topology

Retinal Artery/Vein Classification Based on Multi-Scale Category Fusion

Simultaneous Artery-vein Segmentation in Fundus Images Using an Improved Fully Convolutional Network

Retinal Artery/Vein Classification Based on Multi-Scale Category Fusion.

Automatic Vessel Segmentation on Retinal Images

Artery-venous classification in fluorescein angiograms based on region growing with sequential and structural features

Retinal Vascular Network Topology Reconstruction and Artery/Vein Classification via Dominant Set Clustering

A Review of Retinal Vessel Segmentation and Artery/Vein Classification

Automatic Detection Method of Blood Vessel for Color Retina Fundus Images

Automatic Artery/Vein Classification Using a Vessel-Constraint Network for Multicenter Fundus Images