A New Analysis Method for Retinal Oximetry to Improve Repeatability of Optical Density Ratios

Abstract

Purpose:: Determination of hemoglobin (Hb) O₂ saturation (SO₂) depends on optical density ratio (ODR) calculation and this parameter is variable along the length of retinal vessels and is susceptible to error caused by variable imaging conditions. An analysis method improving repeatability is presented.

Methods:: Oximetry images were obtained in 7 subjects on two occasions, three months apart. Several anatomical locations were analyzed for each image for a total of 54 cases (n=54) and three repetitions were performed at each point. Images contained oxygen-sensitive 605 nm and oxygen-insensitive 586 nm (isosbestic) (±2.5nm) light acquired simultaneously with a 16-bit camera. To account for camera noise, a black image was subtracted from the retinal image. The 605nm and 586nm sub-images were aligned to a high degree of registration using random artifacts as landmarks. The user identified the vessel and specified area for analysis; customized software defined two measurement windows, one on the vessel and one on nearby tissue, both proportional in size to vessel diameter; then, the software identified minimum (darkest) pixel values in the vessel window and maximum (brightest) pixel values in the tissue window. Pixel values were averaged and variation of values was monitored to avoid artifacts. Selection avoids the inclusion of reflected light that has not interacted with Hb while maximal pixel values provide an estimate of maximal reflectance. Software ensures identical placement of analysis windows in both sub-images. Optical density (OD) was calculated as follows: OD=ln(tissue avg/vessel avg). ODs were calculated at 5nm bandwidth wavelengths centered at 605nm and 586nm. ODR, which is proportional to Hb SO₂, was defined as OD₆₀₅/OD₅₈₆.

Results:: ODR repeatability was assessed by the Coefficient of Variability (CoV; standard deviation/mean) of successive measurements. CoV was 4.4% for all locations. Repeatability was better for veins (CoV=1.4%) than arteries (CoV=7.3%) and improved in anatomical locations with reproducibility < 10% (CoV=1.5%).

Conclusions:: The new analysis method yields improved repeatability for ODR values obtained in retinal vessels. The results suggest that locations with high reproducibility better characterize the metabolic state of the retina. This method enhances our ability to investigate retinal metabolism and understand its role in ocular disease.

Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • metabolism • image processing