Automated Measurement of Oxygen Saturation in Retinal Blood Vessels with Visible-light OCT

Abstract

Purpose : To automatically determine oxygen saturation (sO₂) in rat retina objectively and reliably using visible-light optical coherence tomography (vis-OCT).

Methods : A 1.7-µm axial resolution, fiber-based vis-OCT system centered at 560 nm was built and used to image the retina of brown Norway rats (17 weeks old). Volumetric scans were acquired over 2.2 × 2.2 mm² disc regions by using 3 repeated B-scans at each of 512 raster positions; each B-scan comprised of 512 A-lines. Posterior borders (bottoms) of retinal arteries and veins were automatically detected on structural OCT. Bottom pixels were spectroscopically analyzed to measure hemoglobin absorption and calculate sO₂. Circular Doppler scans (d=0.9 mm) consisting of 4096 A-lines were also acquired around disc. Doppler phase shift were calculated between adjacent A-lines. Vessels identified as arteries and veins by oximetry using vis-OCT were compared to those from fundus images and Doppler OCT images (Fig. 1). The oxygen concentration in the inhalation gas was decreased from 100% to 50% and to 21%, and then back to 50% and finally 100%. Arterial sO₂ (A-sO₂) and venous sO₂ (V-sO₂) were measured and oxygen extraction (A-V sO₂) were derived at the different inhaled oxygen concentrations. All experiments were repeated 4 weeks later to test inter-session reproducibility.

Results : One eye of each rat was randomly selected and six eyes were imaged. Arteries, with a higher sO₂ by vis-OCT oximetry, could be distinguished from veins. This arteriovenous distinction agreed Doppler OCT flow direction and with morphologic appearance in fundus images (Fig. 1). A-sO₂ measured by vis-OCT mirrored systemic arterial oxy-hemoglobin saturation (SaO₂) as determined by pulse oximetry (correlation: 0.92, p-value<0.001). In hyperoxia (100% O₂) (Fig. 2 and Table 1), A-sO₂ was 99% ± 1% (mean ± population standard deviation) and the V-sO₂ was 87% ± 2%. In normoxia (21% O₂), A-sO₂ dropped to 92% and V-sO₂ dropped to 69%, suggesting an increased oxygen extraction from 12% to 23%. A-sO₂ and V-sO₂ returned to original levels when the inhalation gas was returned to 100%. The inter-session reproducibility (pooled standard deviation) of A-sO₂ and V-sO₂ was good at all 3 conditions (Table 1).

Conclusions : Automated oximetry based on vis-OCT reliably differentiates oxygen saturation between arteries and veins and detects a greater tissue oxygen extraction at normoxia than that at hyperoxia.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.