Automatisation and improved repeatability of retinal oximetry

Abstract

PurposeTo develop and test an automatic method for analysing retinal oximetry images.MethodsThe dual‐wavelength Oxymap T1 oximeter captures images for calculation of oxygen saturation in retinal vessels. Images were taken of healthy subjects for testing of repeatability and sensitivity, the latter during 100% O2 breathing (n = 21). Images were also taken of patients with diabetes and no DR (n = 16) and patients with non‐proliferative DR (n = 24). The diabetic groups were compared with 54 healthy subjects.Semi‐automatic and fully automatic computer programs were used for analysis. The main differences are: The automatic program automatically categorises vessel points into arterioles and venules and then gives averages for each retinal image. The automatic program uses all pixels on vessel cross‐sections for calculations.ResultsFor arterioles in healthy, the standard deviation between repeated measurements was 0.98 percentage points (pp) with the semi‐automatic method but 0.66 pp with the automatic method. The corresponding values for venules were 1.99 pp (semi‐auto) and 1.50 pp (auto). The automatic program detected an increase of 5.1 pp in arteriolar saturation and 17.4 pp in venules when healthy subjects inhaled pure oxygen (p < 0.0001 in both cases). Similar values were found with the semi‐automatic software.Higher venous saturation was detected in patients with non‐proliferative DR (p < 0.05 with both programs). Values (mean ± SD) from the automatic software were 68.5 ± 5.6% (NPDR), 63.3 ± 6.0% (DM but no DR) and 64.9 ± 4.7% (healthy).ConclusionsAutomatic analysis of retinal oximetry images yields good repeatability. The automatic method is sensitive to changes in saturation during 100% O2 breathing and detects higher venous saturation in diabetic retinopathy. Automatic analysis is much faster than the semi‐automatic approach and is less subjective.