Automatic Retinal Oximetry

Abstract

Abstract: : Purpose: To develop and test new software to automate retinal oximetry and improve its reproducibility and sensitivity. Methods: The software automatically identifies the optic disc and retinal blood vessels on fundus images, which are obtained with four different wavelengths of light. It selects measurement points on retinal arterioles and venules as well as reference points on the adjacent fundus. For comparison, measurement points were also selected manually. The selected points are used to calculate optical density ratios (ODR) which are inversely related to hemoglobin oxygen saturation. Reproducibility was evaluated by measuring the same vessel on five consecutive images. The images were taken of diabetics with and without retinopathy and of healthy volunteers. Sensitivity was evaluated by comparing automatic measurements on healthy volunteers which inhaled either room air, 100% oxygen or 12% oxygen. Results: When arterioles were measured manually, the standard deviation between five images of the same eye was 46% (15–132%) of the mean ODR (mean and range, n=11). When the same vessels were measured automatically the standard deviation was 15% (7–25%). The standard deviations for venules were 18% (4–77%) when measured manually and 8% (3–15%) when measured automatically. During systemic hyperoxia (99–100% O₂ saturation, n=5), ODR in venules was 0.159 (0.109 to 0.209) lower than during normoxia (mean and 95% confidence interval). Conclusions: The automatic computer algorithm improves the reproducibility of retinal oximetry compared to the manual approach. The system can detect changes in retinal oxygen saturation when subjects inhale different concentrations of oxygen.

Keywords: imaging/image analysis: clinical • clinical (human) or epidemiologic studies: systems/equipment/techniques • hypoxia