Dual–Wavelength Retinal Oximetry Using Automated Vessel Tracking and Reconstruction of the Vascular Tree

Abstract

Abstract: : Purpose: To develop an integrated method to automate vessel oximetry from dual–wavelength retinal images, combining methods for automatic vessel tracing, reconstruction of the vascular tree, and measuring the blood oxygen saturation over each segment. Methods: Images are obtained at 570 nm and 600 nm and the vasculature is traced from the 570 nm image using a fully automatic exploratory method. Starting from points on an evenly spaced grid, the algorithm estimates vessel centerlines, thickness, directions, and locations of landmarks such as bifurcations and crossover points. The hierarchical structure of the vascular network is then recovered from the trace fragments and landmark using an algorithm which starts from the prominent vessels and recursively finds the fragments associated with this vessel using connectivity and directionality constraints of the vessels. Fragments belonging to the same segment are grouped together and given the same name while daughter vessels are given a name reflecting the hierarchical structure. Optical densities (OD₅₇₀) are measured for each segment using the minimum reflected intensities inside the vessel and the average outside intensities along each vessel cross section, and organized based on its level in the vascular tree. The 600 nm image is registered to the 570nm image and the OD₆₀₀ measurements for the corresponding vessels are obtained from the 600nm image. Results: Data from 6 subjects was processed using the vessel tracing and linking algorithm. Measurements were made on the prominent veins, arteries and their daughter vessels. The OD ratio (ODR =OD₆₀₀/OD₅₇₀) bore an inverse relationship to systemic HbO₂ saturation (SO₂) providing a powerful indicator of SO₂ variations. The sensitivity to oxygen change found with this automated method was 0.0226 ODR units compared to 0.0236 ODR units previously determined by dual–wavelength oximetry using a method that required much greater user intervention. Conclusions: An automated system for retinal vessel oximetry has been developed. The present experiments demonstrate that oxygen changes are forthcoming from segments of prominent vessels, including arteries and veins in dual–wavelength images. A fully automated system for retinal vessel oximetry will enable early assessments of risk for progression of disease conditions associated with oxygen utilization.

Keywords: imaging/image analysis: clinical • diabetic retinopathy • hypoxia